<![CDATA[High-Impact Low-Probability (HILP) events such as volcanic-collapse tsunamis are difficult to forecast due to their rarity and catastrophic impacts. The December 22, 2018 Anak Krakatau flank collapse illustrated how medium-scale landslides can generate destructive tsunamis without seismic precursors. This study evaluates maximum tsunami amplitude forecasts using 320 pre-computed collapse scenarios. Results show that most cases produced moderate waves (mean 1.17 m), while only 1.9% exceeded 3 m and were classified as extreme. Neural Network models achieved high accuracy for moderate cases but failed to capture outliers representing rare extremes. These findings confirm the structural limitations of scenario-based databases in representing the statistical tails of HILP events. As a preliminary study, this work emphasizes the need for novel algorithms, such as importance sampling, subset simulation, and physics-informed generative models, to improve rare event representation and strengthen volcanic tsunami early warning systems.]]>
Copyrights © 2026
