Towards Slow Earthquakes Forecasting

Slow Slip Events (SSEs) are episodic slip events that play a significant role in the moment budget along subduction megathrust. They share many similarities with regular earthquakes, and have been observed in major subduction regions like, for example, Cascadia, Japan, Mexico, New Zealand. They show striking regularity, suggesting that it might be possible to forecast their size and timing, but the prediction of their extension and exact timing is still yet to come. They certainly are a great natural system to study how friction works at scale of the order of hundreds or thousands of km, and their recurrence time being much shorter than that of regular earthquakes, they give us the possibility to study multiple cycles and test their predictability.
Here we focus on the Cascadia region, where SSEs recur every about 1 or 2 years, depending on the latitude. The study of GPS position time series during the time span ranging from 2007 to 2017 has revealed a low-dimensional (< 5) non-linear chaotic dynamics with a predictable horizon (calculated as the inverse of the metric entropy) in the order of days to months for causally filtered data. It is notable that the increase of instantaneous dimensionality of the attractor seems to constitute a reliable precursor of the large SSEs. The causal filter adopted to reach this conclusion introduces a group delay larger than the predictability horizon time, meaning that this approach cannot be used for real-time forecasting. We thus test alternative filters and data driven approaches (e.g., dynamic mode decomposition) for real-time characterization of the attractor’s properties and evolution. In any case, we conclude that SSEs in Cascadia can be described as a deterministic, albeit chaotic, system rather than as a random process. As SSEs might be regarded as earthquakes in slow motion, regular earthquakes might be similarly chaotic and predictable for short amount of times. If the relation between predictability horizon and the duration of the instability (i.e., slipping event duration) holds also for regular earthquakes, this would imply that earthquakes long-term predictions are intrinsically impossible, and the predictable horizon would be only a fraction of the regular earthquakes typical duration (10-100 s for M>6 earthquakes).