Temblor developed its model by building what we believe to be the largest worldwide ground motion dataset every assembled. We then used a training set of the earthquakes with the best recorded shaking in a Machine Learning process to find a model as a function of earthquake magnitude, depth, and site amplification from our SiteAmp model, which also affects the shaking.

We then compared observed and modeled shaking from twenty key M 6-9 shocks, as well as 500 M≤6 shocks, with representative subsets on continental transform faults (such as the San Andreas and East Anatolian Fault zones), continental thrust faults (such as the Longmenshan fault, site of the 2008 Wenchuan shock), interface subduction quakes (such as on the Japan Trench), and intraslab extensional and compressional quakes (these strike within the subducting slab). Distance to the ground motion stations is measured from the rupture trace except in subduction zones, where it is measured from the rupture periphery.

The 2023 M 7.8 Kahramanmaraş earthquake (below) is the best recorded large event the world has ever known, particularly close to the fault rupture. Even though the near-faut shaking reached unprecedented levels for an earthquake of its size, the Temblor model captures the observed shaking very well.