Abstract Slow, aseismic fault slip has emerged as a significant contributor to the seismic cycle. However, whether slow and fast slip arise from similar physical processes remains unresolved, due to detection biases affecting noisy surface measurements and the analysis of the source properties of slow slip. Using daily geodetic time series denoised with a deep learning model, we invert for 15 years of slow slip evolution on the Cascadia subduction with unprecedented temporal resolution. Our observations show that an upper bound for slow‐slip moment rates exists, and that scaling laws are strongly influenced by the chosen detection threshold and the signal‐to‐noise ratio. Moment rate functions evolve with magnitude: slow slip nucleates as a two‐dimensional expanding crack, propagating laterally when encountering the along‐dip limits of the transition zone. Our findings highlight a continuum of slow slip events of various sizes controlled by subduction interface geometrical constraints.
Costantino et al. (Tue,) studied this question.