Surrogate model approaches – providing fast-running approximation of complex dynamical models – have been gaining traction in the geoscience community. They are highly efficient for uncertainty quantification, and, once trained for a given location, can be used without additional numerical resources and can be seen as a natural solution for PTHA in regions of the world where observations and modelling capacities are lacking. This recipe explores the application of surrogate modeling approaches to two different cases: (1) tsunami hazards in the northern Cascadia, and (2) meteotsunami hazards in the Adriatic Sea.