← Back to Sessions

Correlations of Deep Low-Frequency and Crustal Earthquake Activity in Parkfield, Ca, and Implications for Their Joint Use in Forecasting Frameworks

Description: 

Faults exhibit a wide spectrum of slow slip behaviors that shape the distribution of forces on a fault network, thus informing when and where seismicity occurs. Although it is well documented that large (M>6) slow slip transients can both trigger seismicity swarms and be influential in the nucleation of larger earthquakes, the contribution of smaller-magnitude (M<5) slow slip events to shaping nearby seismicity is still elusive. At the edge of the creeping section of the San Andreas fault near Parkfield, deep slow-slip events are small enough that they are mostly detected indirectly through the activity of low-frequency earthquakes (LFEs). Using them as a proxy for deep slip, our objective is to understand the inter-dependence of deep slow-slip and seismicity in the upper crust, to assess if LFEs could help forecast shallow seismicity changes.

In this work, we show that a significant increase in seismicity occurs both before and after days of high LFE activity. It is consistent with a systematic interaction of deep slow-slip and shallow seismic release. The signal we detect may very well be a sign that the two catalogs are correlated for other reasons. We therefore evaluate possible external sources of correlation, including dynamic triggering of both populations from distant earthquakes and inter-dependent detection thresholds for both catalogs. Such effects would systematically bias any seismicity forecast using LFEs, by interpreting activity correlation as causal interaction between the populations during the learning phase. We finally assess the added value of incorporating LFE activity as additional information to predict seismicity using a deep-learning forecasting model (RECAST, Dascher-Cousineau et al., 2023). The model learns how to predict the timing of the next event based on timing and magnitude and is flexible enough to integrate additional observables to describe the history of activity. We therefore train the model to learn from the number of LFEs in a short period before each earthquake, evaluate the change in forecasting performance and discuss the added-value to the forecasting problem.

Session: New Insights into the Development, Testing and Communication of Seismicity Forecasts [Poster Session]

Type: Poster

Date: 5/2/2024

Presentation Time: 08:00 AM (local time)

Presenting Author: Gaspard

Student Presenter: No

Invited Presentation: 

Authors