The influence of seismic source heterogeneity and local site effects on intensity attenuation during Chilean megathrust earthquakes

Intensity Prediction Equations (IPEs) for Chilean megathrust earthquakes have traditionally relied on simplified point-source representations, such as hypocentral distance ( R h y p ), often neglecting source heterogeneity and local site effects. This study evaluates the performance of alternative distance metrics based on megathrust heterogeneity (i.e., asperities). We define and test three asperity-based metrics: (i) the distance to the nearest asperity ( R a s p ), (ii) the distance to the asperity with maximum coseismic slip ( R a s p m a x ), and (iii) a slip-weighted asperity distance ( R a s p p o n d ). Using a comprehensive dataset of MSK-64 macroseismic intensities from seven major earthquakes with moment magnitudes greater than 8.0 ( M w > 8 . 0 ) over the last 300 years, we calibrated and validated these metrics using a mixed-effects regression approach. Our results consistently demonstrate that asperity-based metrics significantly reduce intensity prediction uncertainty and errors, with improvements of up to 30% compared to R h y p . Specifically, R a s p m a x provides the best fit for large instrumental events, while R a s p p o n d effectively captures the macroseismic fields of historical ruptures. Additionally, incorporating the soil fundamental frequency ( f 0 ) as a proxy for site response was found to be essential for mitigating systematic bias in frequency-dependent residuals. The strong correlation between interseismic locking and coseismic slip could be use to estimate a priori asperity-based metrics to generate different hazard scenarios. These findings underscore the necessity of accounting for source and site complexity to improve seismic risk assessment and territorial planning along the Chilean subduction margin. • MSK-64 intensity models improved for Chilean megathrust earthquakes. • Source heterogeneity reduces intensity prediction error. • Site effects inclusion reduces model intra-event variability and bias.