Distributions of slope angles in tectonically active mountain belts point to the development of threshold conditions, where hillslopes attain a critical inclination or height at which they fail readily because of limitations in material strength. It has been proposed that hillslopes adjust to rapid uplift and bedrock incision through an increase in the rate of relief-limiting landsliding rather than gradual slope steepening. Here we test this concept by investigating the relationship between mean local relief H¯, which we take to be a proxy of long-term erosion rates E, and the occurrence of over 300 of the largest (V > 10⁸ m³) terrestrial landslides on Earth. We find that nearly two-thirds of these giant landslides have occurred in the steepest 5% of the Earth's land surface, where relief is close to its proposed upper strength limit. They are primarily located in deeply incised valleys, along fault-bounded fringes of active mountain belts, and in volcanic arcs.
This distribution coincides with areas of high long-term erosion rates (∼ 4 mm yr^-1), confirming that giant landslides contribute to rapid denudation of mountains. Most of the eroded volume is concentrated in the smallest, but steepest parts of mountain belts and volcanic arcs. First-order estimates of minimum erosion rates accomplished by the largest landslides are ≥ 0.01 mm yr^-1 these rates are between 1% and 10% of the Late Pleistocene to Holocene mean erosion rates in a given area. Importantly, the landslide erosion rates show a nonlinear increase with mean local relief, suggesting that the contribution of giant landslides in total and per event increases significantly with increasing overall erosion rates. However, giant landslides also occur in areas of lower-than-average relief (H¯ ∼ 300-700 m), irrespective of whether threshold hillslopes have developed or not. Factors contributing to these failures include soft rocks, extensive low-angle discontinuities, high rates of fluvial bedrock incision, and tectonically driven deformation and slope loading.