Precursory tremor of the Askja Caldera landslide, July 2014 - seismic signal analysis and numerical modelling

Seismic records can contain valuable information about triggers and precursors of slope failures that might become useful for early-warning purposes. We investigated the seismic data of 52 stations from the University of Cambridge, UK, with respect to the tremor signals preceding a 20-80x106 m3 landslide at the Askja caldera in the Icelandic highlands on 21 July 2014. The landslide created a tsunami in the caldera lake, which inundated the shore up to 60 m high reaching famous tourist spots. This shows the high hazard potential of the site that motivated this study. About 30 min before the landslide, the seismic ground velocities >1 Hz of stations up to 30 km away from the landslide source area started to increase and the tremor signal reached up to three times the background noise level about 7 min before the landslide. In the spectral domain, the tremor is visible as a continuous, harmonic signal with a fundamental frequency of 2.5 Hz and overtones at 5 and 7.5 Hz. About 10 min before the landslide, the activated frequency bands changed their spectral content and up and down gliding is observed contemporaneously. The tremor signal ceases about 5 min before the high-energy failure of the landslide. We interpret the harmonic tremor before the landslide as stick-slip motion on fault patches at the boundaries of the landslide mass. Individual stick-slip events cannot be distinguished in the seismic data and thus have already merged into continuous tremor as they occur very close in time. As up and down gliding of the frequency bands occurs at the same time we favour an explanation where several fault patches are active simultaneously. One patch might accelerate and create up gliding signals and another patch might decelerate and create down gliding. We matched synthetic seismograms produced by numerical simulations of stick-slip movement and the seismic observations. The results show that a patch with a radius of 45 m and a realistic landslide thickness of 30 m can reproduce the spectrograms from observations. The disappearance of tremor shortly before the landslide is consistent with the theoretical prediction of a transition to stable sliding at high loading rates, which could imply that the landslide mass had already started to move before the high-energy signals are emergent in the seismic data.