Earthquake hazard forecasting often has to deal with the scarcity of seismicity data, the complexity of fault segmentation, the uncertainties derived from the characterization of seismogenic sources, the dynamics of earthquake rupture and the recurrence model used. Instrumental earthquake recordings are available only since about 1900 and they are not reliable until several decades later. There are a few regions with representative historical data of large earthquakes for more than 2000 years in Italy (Boschi, 2000), Japan (Ishibashi, 2004), and China (Liu et al., 2011) that allow acquisition of reliable major earthquake magnitudes and recurrence statistics. In regions with low to moderate seismicity, such as some zones of the western Mediterranean or intracontinental areas, earthquake hazard estimation is especially challenging (see e.g., Estay et al., 2016;Gamage & Venkatesan, 2019;Perea & Atakan, 2007). Most of the large, and consequently damaging earthquakes, have recurrence times on the order of hundreds or thousands of years. Since historical catalogs are often scarce and imprecise, the paleoseismic record provides valuable data about rare but devastating major events and long-term recurrence times, but these data usually have large uncertainties (McCalpin, 2009). Often, the slow fault systems, besides generating low seismicity in terms of frequency, show little geomorphological expression. This makes it difficult to define the segmentation of fault traces and the parameters relating to their 3D geometry and their kinematics. Fault system geometry also plays an important role in the stress interactions that control the earthquake ruptures and therefore magnitude and recurrence statistics (e.g.