[S-CG45] Science of slow-to-fast earthquakes

[SCG45-P40] Data assimilation for reproducing and predicting the fault slip behavior in the 2010 Bungo Channel long-term slow slip event

*Masayuki Kano^1,2, Yusuke Tanaka¹, Takeshi Iinuma², Takane Hori² (1.Graduate school of science, Tohoku University, 2.JAMSTEC)

Keywords:Slow Slip Event, Data Assimilation, Frictional Properties, GNSS

Data assimilation (DA) is the technique to combine the observations and physics-based simulations. DA has now been widely adopted in the field of meteorology and oceanology, especially in its practical use such as weather forecast. It has recently been applied to the problem of fault slip estimation. For example, Kano et al. (2020) developed an adjoint DA method and applied to the postseismic crustal deformation data following the 2003 Tokachi-oki earthquake. They assimilated GNSS data for 15 days following the mainshock to optimize frictional properties of the afterslip area, and then examined the short-term predictability of GNSS data for the following 15 days. Hirahara and Nishikiori (2019, hereafter, HN19) proposed Ensemble Kalman Filter method and investigated the feasibility for estimating the fault slips during slow slip events (SSEs) through numerical experiments.
Following these studies, we attempt to assimilate GNSS data including long-term SSE (LSSE) in the Bungo Channel, southwest Japan, occurred during 2009-2011.We used the same fault model as HN19 covering the Bungo Channel LSSE area, consisting of one large circular patch allowing for the occurrence of SSE within the surrounding stable sliding region. By assigning the conditionally stable frictional properties, HN19 reproduced recurrent LSSEs with a similar recurrence interval, duration, and maximum slip velocity observed in the Bungo Channel LSSEs. Assuming these frictional properties as the initial model, we attempt to optimize the frictional properties in the LSSE patch by DA, and discuss the reproducibility and predictability of GNSS data.