← Back to Sessions

Evaluating the Forecasting Performance of U.S. Geological Survey Aftershock Forecasts

Description: 

The U.S. Geological Survey (USGS) releases public aftershock forecasts following earthquakes with magnitudes 5 or larger in the U.S. and related territories. This system provides forecasts for specific durations (next day, week, month, year) and magnitude thresholds (M3+, M4+, …, M7+). We release the forecasted distribution for the number of aftershocks in each duration-magnitude bin, represented by a fine grid of quantiles. Forecasts are produced soon (20 minutes) after the mainshock and get updated regularly in intervals that increase with time. We are prototyping a forecasting system that uses the temporal Epidemic-Type Aftershock Sequence (ETAS) model and also accounts for short-term aftershock incompleteness using a parametric model. Early in the sequence, ETAS parameters are drawn from generic distributions of values that have fit previous aftershock sequences in that tectonic region, and as the sequence develops, parameter distributions are updated in a Bayesian manner to fit the sequence specifically. In this work, we evaluate the performance of this proposed aftershock forecasting system, by comparing the forecasted distribution of aftershock counts in each forecast bin to the observed count, using the USGS’s Comprehensive Catalog. We employ a battery of forecast metrics designed for such probabilistic comparisons and that are meaningful for the forecasts’ end-users. In specific, we consider metrics that can reveal (1) the scale of misprediction of the forecasted distribution, (2) the precision of the forecasted distribution and (3) the tendency of the forecasted distribution to over- or under-predict. We consider whether and how to account for aftershock incompleteness and whether and how to perform sequence-specific parameter estimation for different ETAS parameters and the effects of these decisions on forecast performance. We also examine forecast performance across the forecasts’ dimensions: forecast duration, magnitude threshold, and time since mainshock. We present first findings of our evaluation strategy applied to simulated aftershock forecasts made under this proposed system.

Session: Improving the State of the Art of Earthquake Forecasting Through Models, Testing and Communication [Poster]

Type: Poster

Date: 4/15/2025

Presentation Time: 08:00 AM (local time)

Presenting Author: Jeanne

Student Presenter: No

Invited Presentation: 

Poster Number: 136

Authors