Automated workflow to compute earthquake chronologies on faults from paleoseismic datasets

A major challenge in seismic hazard research is to quantify the frequency of large earthquakes along active faults, more so when the observational time windows of seismic catalogs are much shorter than the average fault recurrence intervals. In this respect, paleoseismology continues to prove to be an excellent tool to extend the seismic catalogs of faults into prehistorical times. The combination of the ever more advanced trenching surveys and accurate numerical dating techniques allows constraining the timing of paleoearthquakes and, for some datasets, approximating their recurrence models. Despite this, paleoseismic data carries along large uncertainties frequently related to dating technique limitations, poor stratigraphic preservation, and along-strike slip variability that hinder the identification of a complete paleoearthquake record. Subsequently, these issues, among others, challenge the constraint of reliable earthquake chronologies along faults and of the parameters defining their earthquake cycle.We present an automatic workflow capable to compute and constrain earthquake chronologies along a fault based on the correlation of its available paleoseismic records, including multi-site and poorly constrained datasets. Our inherent premise is that the correlation of paleoseismic data from multiple along-fault locations can help to improve the time constraints and completeness of its paleoearthquake record. Given that paleoseismic records are, by definition, underpopulated, event correlation is not restricted to single occurrences. Instead, an event in a site might be simultaneously correlated with more than one in another if time compatible. Furthermore, to avoid subjectivity biases in event timing estimates and correlation, we exclusively rely on the trench numerical dates limiting each event horizon as the inputs. All earthquake chronologies are modelled probabilistically with a four-step algorithm as we detail. First, all earthquake times in each site are computed as probability density functions (PDFs) using the input numerical dates. The event PDFs from all sites are then integrated to derive a mean curve representing the overall event probabilities for the studied fault in the time span investigated. The probability peaks in this curve, which are assumed as indicative of the event timing at the fault scale, are automatically detected based on peak prominence analysis. A final PDF is then computed for each peak by multiplying all site event PDFs intersecting the peak position. The set of product PDFs constitutes the earthquake chronology of the fault, provided to the user in simple output files that can be externally used to calculate fault parameters for the seismic hazard assessment, and to visualize the modelling.Preliminary tests on several paleoseismic datasets of the Central Apennines (Italy), the Eastern Betics (Spain), the Dead Sea Fault and the Wasatch Fault (US), have provided good outcomes. The approach significantly reduces the uncertainties in event timing of paleoearthquakes and provides an objective and reliable interpretation of the datasets, especially when these are complex or have wide uncertainties. By extension, the workflow has the potential to reduce the uncertainties in earthquake recurrence estimates and can give insight on the recurrence models that better describe the earthquake cycle in the studied faults.