Rapid and accurate hazard forecasting is important for prompt evacuations and reducing casualties during natural disasters. In the decade since the 2011 Tohoku tsunami, various tsunami forecasting methods using real-time data have been proposed. However, rapid and accurate tsunami inundation forecasting in coastal areas remains challenging. Here, we propose a tsunami forecasting approach using convolutional neural networks (CNNs) for early warning. Numerical tsunami forecasting experiments for Tohoku demonstrated excellent performance with average maximum tsunami amplitude and tsunami arrival time forecasting errors of ~0.4 m and ~48 s, respectively, for 1,000 unknown synthetic tsunami scenarios. Our forecasting approach required only 0.004 s on average using a single CPU node. Moreover, the CNN trained on only synthetic tsunami scenarios provided reasonable inundation forecasts using actual observation data from the 2011 event, even with noisy inputs. These results verify the feasibility of AI-enabled tsunami forecasting for providing rapid and accurate early warnings.

The 2011 Great East Japan Tsunami exposed many hidden weaknesses in Japan’s tsunami countermeasures. Since then, many improvements have been made in both structural measures (numerical simulations, coastal defense structures, building damage assessment and control forests) and nonstructural measures (warning/observation and evacuation). This review summarizes the lessons and improvements in the five-year time period after the 2011 event. After five years, most of the lessons from the 2011 tsunami have been applied, including more realistic tsunami simulations using very fine grids, methods to strengthen coastal defense structures, building evacuations and coastal forests, improved warning content and key points to improve evacuation measures. Nevertheless, large future challenges remain, such as an advanced simulation technique and system for real-time hazard and risk prediction, implementation of coastal defense structures/multilayer countermeasures and encouraging evacuation. In addition, among papers presented at the coastal engineering conference in Japan, the proportion of tsunami-related research in Japan increased from 15% to 35% because of the 2011 tsunami, and approximately 65–70% of tsunami-related studies involve numerical simulation, coastal structures and building damage. These results show the impact of the 2011 tsunami on coastal engineering related to academic institutions and consulting industries in Japan as well as the interest in each tsunami countermeasure.

Rainfall-induced landslides cause various types of damage, including damage to infrastructure, leading to devastating economic losses and human casualties. Although various numerical methods have been developed to predict landslide occurrence and the extent of sediment flow, three-dimensional analysis of the entire landslide process in a unified manner is still challenging owing to limitations in computational efficiency and the representation of deformation and flow. In this study, we present a numerical method of rainfall-induced landslides using the coupled hydromechanical material point method (MPM) for unsaturated porous media based on implicit and explicit formulations, which enables the efficient analysis of all rainfall-induced landslide processes, including both quasi-static and dynamic processes. The developed method includes an implicit MPM based on a simplified formulation, which is first applied to the quasi-static analysis in the pre-failure stages in which rainwater infiltrates the ground. The explicit MPM is then applied to the dynamic analysis for post-failure stages in which the ground collapses and flows. A constitutive law for soils is improved in the simulation of landslide initiation and sediment flow by incorporating the effect of cohesion in a visco-plastic model for granular materials. The proposed method was applied to a three-dimensional terrain model of Ashikita town, Kumamoto, Japan, where an actual landslide occurred owing to intense rainfall on July 3–4, 2020. Numerical results from the proposed hydromechanical coupling compared with single-phase MPM indicated that pore water plays an important role in understanding all rainfall-induced landslide processes, from landslide initiation to sediment discharge.