Time variation in amplitude-frequency distribution of deep non-volcanic tremors in the Bungo Channel region, southwest Japan

Magnitude-frequency distribution of earthquakes follows the Gutenberg-Richter law. The slope of this law, b value, represents the relative occurrence of large and small earthquakes. Since magnitude is defined as corrected logarithmic amplitude, amplitude-frequency distribution of earthquakes is linear on log-log graph. On the other hand, that of non-volcanic tremors is linear on semi-log graph, indicating that it follows the exponential distribution, not the power-law distribution [Hiramatsu et al., 2008]. Thus the slope of amplitude-frequency distribution for tremors is equivalent to b value for earthquakes. In this study, we investigated time variation in the slope of amplitude-frequency distribution from analyses of tremor activities in the Bungo Channel region, where long-term slow slip events occurred in 1997 and 2003. We analyzed vertical-component records from Hi-net and the seismic networks of Japan Meteorological Agency and universities for the five-year period between 2004 and 2008. We also used records from the temporal seismic station in Hiburi Island installed by us in the period after Sept. 2004. In the Bungo Channel region, tremor activity occurs with a recurrence interval of approximately two months. We observed a total of 35 activities including small ones in the analysis period. To detect tremors and determine their hypocenters, we used the same software as Automatic Tremor Monitoring System (ATMOS) [Suda et al., in press]. We obtained frequency distribution of reduced displacements (RDs) for each tremor activity. RD is RMS amplitude of ground displacement corrected with hypocentral distance [Aki and Koyanagi, 1981], and it is proportional to seismic moment rate. The observed slopes of RD-frequency distribution for active swarms were approximately constant in the period between 2004 and 2006, but they declined in 2007. At the end of 2008, the slopes decreased bellow half the values between 2004 and 2006. As well as b value, the slope represents the relative occurrence of large and small tremors: the smaller the slope, the more frequently large tremors occur. Therefore, the result shows that large tremors occurred more frequently in 2008 than in the previous period. Schorlemmer et al. [2005] have suggested that high and low b values indicate low and high differential stress states in source region, respectively. Previous studies on time variation in b values have reported that b values tend to decrease toward a main event. Tidal synchronization of tremors indicates that tremor occurrence is very sensitive to stress changes [e.g. Nakata et al., 2008]. We can expect that slope of RD-frequency distribution for tremors is more sensitive to stress state than b value for earthquakes. It is considered that the next long-term slow slip event is about to occur in the Bungo Channel region. The observed decrease in the slope of RD-frequency distribution may indicate increase in differential stress, which will trigger the slow slip.