“…Approximations of results using methods like Green's Function, amplification factors, and reduced complexity models (Molinari et al, 2016;Løvholt et al, 2016;Glimsdal et al, 2019;Gailler et al, 2018;Grzan et al, 2021;Röbke et al, 2021) 3. Hardware and computational improvements like a nesting of grid domains, adaptive variable grid resolution, parallelisation and GPU-based acceleration, coupled multi-scale modelling, exascale codes (LeVeque et al, 2011;Shi et al, 2012;Oishi et al, 2015;Macías et al, 2017;Marras and Mandli, 2021;Folch et al, 2023) 4. Surrogates using statistical emulators and machine learning models (Sarri et al, 2012;Salmanidou et al, 2021;Mulia et al, 2018;Fauzi and Mizutani, 2019;Makinoshima et al, 2021;Fukutani et al, 2023;Mulia et al, 2022) Running the numerical simulations, especially for modelling the tsunami inundation with an NLSWE model takes a lot of time and computational resources, for example accounting for about 13,600 GPU hours in the local PTHA study for Catanai by Gibbons et al (2020), refer Table S5 for runtimes of this study.…”