Exploring improvements on magnitude estimates for paleo-earthquakes based on data type and volume availability

Yajaira De Haro, Allison C. Schiffmaier, & Alba M. Rodriguez Padilla

Submitted September 7, 2025, SCEC Contribution #14841, 2025 SCEC Annual Meeting Poster #TBD

Estimating paleoearthquake magnitudes often uses simple regressions based on displacement or rupture extent. As more data becomes available in the form of additional displacement measurements or rupture extent constraints from new trenches, can paleo-magnitudes be improved by leveraging these additional datasets? We tested whether a neural network could improve accuracy using modern data from 75 surface-rupturing earthquakes. Comparing the neural net’s performance to three standard methods, we down-sampled data to simulate sparse conditions akin to the limited displacement and rupture extent data from trenches and geomorphic offsets, and evaluated model performance as the data was downsampled. Our preliminary analysis shows that the neural net, leveraging information on displacement and rupture length, outperforms other methods until the data has been downsampled 60%. At that point, surface rupture length extent is the best magnitude predictor. Moment-based magnitudes consistently underperform the regressions and the neural net, no matter the data volume considered, suggesting traditional methods should favor regressions over moment-based estimates. Last, we explore application of the methods to geomorphic offset data from the Coyote Creek fault.

Key Words
Earthquake magnitudes, Paleoseismology, Neural network, Displacement, Rupture length, Downsampling, Regression analysis, Magnitude estimation

Citation
De Haro, Y., Schiffmaier, A. C., & Rodriguez Padilla, A. M. (2025, 09). Exploring improvements on magnitude estimates for paleo-earthquakes based on data type and volume availability. Poster Presentation at 2025 SCEC Annual Meeting.

Related Projects & Working Groups
Earthquake Geology