Research — Disaster Geoinformatics Lab. (ReGiD)

Real-time tsunami inundation & damage estimation

Using real-time earthquake information and supercomputing, and in collaboration with industry and universities, we develop technology to estimate tsunami damage and deliver it within roughly 20 minutes of an earthquake. We call this the "10-10-10 Challenge": 10 minutes to estimate the tsunami source, and 10 minutes to compute inundation on a 10 m mesh. The system was developed with the Research Center for Prediction of Earthquakes and Volcanic Eruptions and the Cyberscience Center at Tohoku University, together with NEC and Kokusai Kogyo Co., Ltd. The work received the Japan Resilience Award 2016 (Excellence Award).

Tsunami simulation Inundation modeling Real-time computing Damage mapping Fragility functions

Real-time tsunami inundation and damage prediction on the SX-ACE supercomputer — Real-time tsunami inundation and damage prediction using the SX-ACE supercomputer at Tohoku University.

Big-data analysis for earthquake & tsunami disaster mitigation (JST CREST)

Adopted under JST CREST in 2015, this project brings together disaster simulation, disaster management science, mathematics, and information science. Through large-scale, high-resolution numerical simulation and real-time data assimilation, we are building one of the world's first big-data analysis platforms for disaster mitigation — presenting, in real time, the measures needed to avoid worst-case scenarios and supporting decision-making by national and local governments.

Data fusion Physics-based modeling Data assimilation Disaster scenarios

Controlling the scenario explosion of disaster simulation — Controlling the "scenario explosion" in large-scale disaster simulation.

Increasing urban resilience to large-scale disasters (Japan–Israel)

In partnership with the Hebrew University of Jerusalem, this project builds a simulation platform for large-scale disaster assessment, response planning, and mitigation. The Japanese side performs short-term damage assessment with simulation and remote sensing, while the Israeli side develops agent-based and detailed land-use models integrated in GIS. Together the teams produce an integrated disaster-management tool at a spatial and temporal level of detail not previously available.

Agent-based modeling GIS Damage mapping Disaster scenarios

Collaboration scheme with the Hebrew University of Jerusalem, Israel.

South American tsunami prediction from simulation & remote sensing

Large earthquakes in Central and South America have caused tsunami disasters with Pacific-wide impact, including in Japan. To improve resilience in tsunami-prone Latin American countries, we combine real-time computing with remote sensing to grasp the full extent of damage. Near-field and far-field tsunami impacts are assessed separately, and we developed a GIS-based system for building-damage estimation using real-time tsunami simulation and fragility functions.

Source inversion Tsunami simulation SAR rapid assessment Fragility functions Damage mapping

Wide-area damage assessment for South American countries — Wide-area damage assessment combining real-time tsunami simulation and remote sensing for South American countries.

Tsunami simulation that reproduces building damage

To predict future tsunami damage more accurately, we need to understand tsunami behavior in densely built urban areas, where structures obstruct the flow. We built a simulation technique — a time-evolving synthetic equivalent roughness model — that reproduces how buildings are washed away as a tsunami floods a city, identifying which buildings fail, when, and how many. Applied to the Sendai Plain (Natori River estuary), it reproduced the building damage of the 2011 Tohoku tsunami and matched aerial imagery from the event.

Tsunami simulation Inundation modeling Physics-based modeling Fragility functions

Observed versus estimated washed-away building distribution — Left: washed-away buildings from the MLIT field survey (2011). Right: damage estimated by the time-evolving synthetic equivalent roughness model.

Next-generation tsunami simulation with the Lattice Boltzmann method

Tsunami simulation has advanced rapidly with growing computing power, now reproducing tsunami height within roughly 10% error. Reproducing tsunami run-up into cities traditionally requires demanding 3D CFD models. By developing tsunami simulation based on the Lattice Boltzmann method — a fully explicit scheme well suited to massively parallel computing — we explore the potential to surpass current speed limits and build a next-generation tsunami simulation technique.

Tsunami simulation Inundation modeling Physics-based modeling HPC

Tsunami inundation simulation with the Lattice Boltzmann method — Tsunami inundation simulation using the Lattice Boltzmann method.

Rapid worst-case tsunami estimation from earthquake early warning

Using the earthquake early warning available immediately after an event, we estimate the worst-case tsunami scenario to support rapid assessment. Underestimation of the tsunami warning during the 2011 Tohoku earthquake delayed evacuation. By setting up many tsunami scenarios and accounting for the uncertainty of initial conditions, our method estimates — in about three minutes — the maximum wave height that may occur in the worst case, with an average accuracy of about 97%.

Source inversion Tsunami simulation Disaster scenarios Earthquake early warning

Assumed fault arrangements for worst-case tsunami estimation — Example fault arrangements, varying parameters such as source orientation, to identify scenarios producing high tsunami amplitude.

Earthquake building-damage assessment with polarimetric SAR

This work assesses earthquake-induced building damage from a single post-seismic dual-polarimetric ALOS-2/PALSAR-2 image. The dataset is well suited to distinguishing slightly damaged, collapsed, and tilted buildings. Using machine-learning algorithms on built-up areas with different damage levels, the method satisfactorily detected building damage in the Kathmandu district following the 2015 Nepal earthquake.

Building damage mapping ALOS-2/PALSAR-2 Deep learning SAR rapid assessment Change detection

Machine-learning building-damage mapping from ALOS-2/PALSAR-2 — Machine-learning method for building-damage mapping using ALOS-2/PALSAR-2 SAR imagery.

Extracting flooded urban areas from L-band SAR

This research detects flooded areas within urban environments using SAR imagery, with the September 2015 Kanto–Tohoku heavy-rain event as a case study.

Flood mapping Change detection ALOS-2/PALSAR-2 SAR rapid assessment

Flooded-area extraction near overtopping locations (left: average difference; right: correlation coefficient).