Effective regional prediction of earthquake-induced landslides: The Site-Adaptable Newmark Displacement (SAND) approach

 Earthquake-trigerred landslides (ETLs) cause a significant portion of total earthquake losses in mountainous regions, threatening both financial stability and community sustainability. For nearly 60 years, the Newmark displacement (ND) method has been widely used to estimate earthquake-induced slope deformation. However, most existing ND models are based on regressions developed from specific earthquakes or datasets, which limit their applicability across different tectonic and climatic settings.

To address this gap, we introduce the Site-Adaptable Newmark Displacement (SAND) approach, a flexible, knowledge- and data-driven method designed to work across a wide range of tectonic environments and spatial scales. SAND assumes a quadratic relationship with peak ground acceleration (PGA) and a non-linear relationship with critical acceleration (A_c) and progressively incorporates regional and site-specific factors such as fault focal mechanisms, hanging-wall and footwall effects, topographic amplification, terrain roughness, and climate-related wetness conditions.

We validated the SAND approach against several catastrophic events, including the 2022 Ms 6.8 Luding earthquake (China), the 2010 and 2021 Haiti earthquakes, and major events in Taiwan (1999) and Lushan (2013, 2022). Our comparative analysis shows that older, site-specific equations, such as Miles and Ho (1999), often outperform newer modified versions that overemphasize slope stability at the expense of seismic intensity attenuation. Specifically, in the Luding case, incorporating slope orientation significantly improved predictive power, accounting for the preferential distribution of landslides on E-, SE-, and S-facing slopes.

Overall, SAND consistently performs better than previous regression-based models (e.g. Jin et al., 2019) in predicting landslide locations. Because this method does not require a pre-existing landslide inventory, it can be implemented immediately following an earthquake using only magnitude, epicentre, and focal mechanism data. This can allow for the rapid prediction of shallow ETLs to support emergency rescue efforts and prioritize resource allocation in high-risk zones.