Aftershock Forecasts Following the M6.4 and M7.1 Ridgecrest, California Earthquakes of July 2019

Session: Observations From the 2019 Ridgecrest Earthquake Sequence [Poster]
Type: Poster
Date: 4/28/2020
Time: 08:00 AM
Room: Ballroom
Description: 

Following the M_W6.4 and M_W7.1 Ridgecrest earthquakes, the USGS issued aftershock forecasts based on the Reasenberg and Jones model (Science, 1989). We had previously updated this method to improve uncertainty estimates, the handling of early catalog incompleteness and adaptation as aftershock sequences progress (Page et al, BSSA 2016). We also developed new regionalized aftershock parameters for California (Hardebeck et al, SRL 2019). The frequently-updated aftershock forecasts for Ridgecrest were integrated with other earthquake products on the USGS event webpages and communicated with a pre-written template that provides basic information as well as the detailed numerical forecasts.

We evaluate the success of the forecasts by comparing them with the observed numbers of aftershocks. Our forecasts give the 95% confidence range of the expected number of aftershocks, and we consider a forecast successful if this range bounds the observed number. There are few M≥5 aftershocks, consistent with our forecasts, so we focus on M≥3 aftershocks for testing. Forecasts following the M6.4 earthquake successfully bound the observed number of aftershocks, until the M7.1 mainshock occurred. Forecasts following the M7.1 mainshock, however, fluctuate between under- and over-prediction, particularly for longer duration (e.g. 1 month) forecasts. The best-fitting p-value, which describes the sequence decay, also fluctuates as the sequence progresses, and does not reach a stable value within the first 4 months. Our hypothesis is that the smooth temporal decay of the Reasenberg and Jones model is too simple for this complex sequence, as it can’t model rate fluctuations due to secondary triggering.

We are planning to implement ETAS forecasting (Ogata, JASA 1988), which models secondary triggering. As a test, we will apply the ETAS model to the same real-time catalogs used for the Ridgecrest forecasts to investigate whether ETAS would have produced more accurate forecasts.

Presenting Author: Jeanne Hardebeck

Authors