Challenging our degree of belief in paleoseismic magnitudes

The estimation of magnitudes from field geological data is a crucial problem in paleoseismology, which is key to reconstructing the history of active faults and to better constrain regional seismic potential. Indeed, coseismic slip and rupture length offer critical clues to infer earthquake magnitude, especially for recently silent faults lacking instrumental or historical records. However, interpreting geological evidence requires caution. Data sampling is non-random and depends on factors such as preservation potential, offset size, and site accessibility. Larger offsets are more likely to be preserved and thus may be overrepresented. Rupture heterogeneity along strike further complicates the use of single-site data to characterize entire events. Therefore, robust methods are needed to incorporate variability and uncertainties in geological data. We present an innovative Bayesian method to estimate paleoearthquake magnitudes by jointly analyzing coseismic rupture length (L), throw (S), and age (T) data from a new, high-quality paleoseismic database for the central Apennines. The approach accounts for different sources of uncertainty using both standard Gaussian models and more flexible, data-driven distributions. We also assess bias due to (1) overestimation of slip values and (2) time-dependent effects. Our method is implemented in soon-to-be-released open-access MATLAB software. We validate the model using dense, high-resolution field datasets from recent Apennine earthquakes such as the 2016 Mw 6.6 Norcia earthquake, comparing our results with classical approaches. We show that while traditional magnitude estimates reported in paleoseismic and parametric catalogs are consistent, their uncertainties are often underestimated even for recent events with abundant freshly collected data with significant implications for seismic hazard assessment.