Preliminary forecast model of crustal earthquakes in southwest Japan based on GNSS data

In Japan, the Headquarters for Earthquake Research Promotion has developed a nationwide probabilistic earthquake model called “National Seismic Hazard Maps for Japan” since the destructive 1995 Kobe earthquake. This model covers both subduction and crustal earthquakes based on a history of past large earthquakes from seismological, archaeological, and geological data. The model for crustal earthquakes relies on geological and geomorphological data of active faults and never use geodetic data, whereas contemporary deformation of the Japanese Islands has been observed by a dense GNSS network. Here, we attempt to develop a preliminary forecast model of shallow crustal earthquakes using GNSS velocity data.

We follow the procedure of Shen et al.(2007) to calculate the forecast model. The GNSS velocities at continuous GNSS stations from April 2005 to December 2009 are used for the model in southwest Japan. Elastic deformation due to interplate coupling along the Nankai Trough is removed using the block model of Nishimura et al. (2018). Strain rate field is calculated at a grid point of 0.2º x 0.2º by a method of Shen et al (1994). The strain rates are converted to geodetic moment rates by a formula proposed in Savage and Simpson (1997). The thickness of a seismogenic layer, rigidity, b value of the Gutenberg-Richter law, and magnitude of the maximum earthquake are assumed to be 12 km, 30 GPa, 0.9, and 7.5, respectively. They are uniform in the modeled region. Previous studies (e.g., Shen-Tu et al., 1994) revealed that geodetic strain rates were much larger than seismological ones in southwest Japan because geodetic strain includes both elastic and inelastic strain. Elastic strain rates presumably equal to seismological ones on a long-term average. We compared seismic moment rates released by shallow historical earthquakes since AD1586 with the geodetic moment rates. Their ratio is 0.24 and 0.16 in the Chubu, Kinki, and Chugoku region and the whole southwest Japan. This difference is probably attributed to the distribution of historical documents and may also reflect the regionality of the ratio between elastic and inelastic strain. Applying 0.16 for calculating elastic rates and the stationary Poisson process of the earthquake occurrence, a probability of M≥6 earthquakes for 30 years ranges from 5.1 % to 0.2 % in each 0.2º x 0.2º grid of southwest Japan. We verify this probability model by using shallow (Depth≤ 20 km) M≥5 earthquakes occurred in 2010-2019, which is a period after the used GNSS data. The number of earthquakes was 36, which is roughly concordant to the predicted number of the model (3.04 per year). About 58 % of the earthquakes occurred with 25 % of the area with the highest strain rates, which suggests many crustal earthquakes occur in high strain-rate regions. The verification suggests the preliminary forecast model has the predictive power reasonably.