Near real-time spatial prediction of earthquake-induced landslides: A novel interpretable self-supervised learning method

Near real-time spatial prediction of earthquake-induced landslides (EQILs) can rapidly forecast the occurrence position widespread just after a violent earthquake; thus, EQIL is very crucial to 72-hour ‘golden window’ for survivors. This work focuses on series earthquake events from 2008 2022 occurring in Tibetan Plateau, famous seismically-active zone, and proposes novel interpretable self-supervised learning (ISeL) method near EQILs. new innovatively introduces swap noise at unsupervised mechanism, which improve generalization performance transferability model, effectively reduce false alarm accuracy through supervised fine-tuning. An module built based self-attention mechanism reveal importance contribution various influencing factors distribution. Experimental results demonstrate that ISeL model superior excellent state-of-the-art machine deep methods. Furthermore, according method, critical controlling triggering are revealed. The also be applied other earthquake-frequent regions worldwide because its good transferability.