Research

Natural disasters — earthquakes, tsunamis, floods, and landslides — cause enormous human and economic losses each year. Our research develops computational tools that harness satellite remote sensing data, machine learning, and physics-based numerical simulation to monitor, model, and rapidly assess the impact of these events. A core theme is the combination of data-driven and model-driven approaches to produce actionable, scalable information for disaster response and planning.

Research Topics

Topic 01

Satellite Remote Sensing & Image Analysis

Here, we develop intelligent methods for automatically extracting disaster-related land information from synthetic aperture radar (SAR) and optical satellite imagery. These methods include multi-temporal change detection and SAR intensity image analysis to map earthquake and tsunami damage, delineate flood inundation, and generate landslide inventories.

Building damage mapping Flood Mapping Change Detection ALOS-2/PALSAR-2 Sentinel-1

Representative papers: Kametaka et al. (2025) Int. J. Applied Earth Obs. Geoinf.; Endo et al. (2018) Remote Sensing; Karimzadeh et al. (2018) Remote Sensing.

Topic 02

Machine Learning & AI for Disaster Science

Here, we formulate and design AI models — neural networks, deep learning, and generative models — for multi-class building damage classification, post-disaster change detection, and data augmentation. A key focus is multimodal learning that jointly leverages SAR, optical, and ancillary geospatial data.

Deep Learning CNNs and Transformers Generative Models Multi-modal Fusion Semantic Understanding

Representative papers: Ho et al. (2025) Remote Sensing; Sriyanto et al. (2015) Ocean Engineering; Ezaki et al. (2024) IEEE GRSL; Adriano et al. (2021) ISPRS J. Photogramm. Remote Sens..

Topic 03

Multi-Hazard Damage Assessment

Here, we develop integrated frameworks for rapid, large-scale mapping of building damage after major disasters worldwide, including earthquakes, tsunamis, floods, and landslides. These frameworks combine image analysis with AI models applied to Earth observation data to produce actionable damage maps within hours of an event.

Damage MappingFragility FunctionsSAR Rapid AssessmentEarthquakeTsunami

Representative papers: Karrel at al. (2025) Prog. Disaster Sci.; Nagato et al. (2025) Journal of Disaster Research; Adriano et al. (2019) Remote Sensing.

Topic 04

Multi-modal Data Fusion & Digital Twins

Here, we explore physics-based modeling and deep learning methods that jointly leverage SAR data, optical imagery, land-surface information, aggregated population data, and simulation outputs to produce structured 3D representations of disaster scenarios, toward digital twins and complex real-world reconstruction.

Data Fusion Physics-based Modeling Bayesian Methods Digital Twins Disaster Scenarios

Representative papers: Liu et al. (2026) Journ. of Disaster Science; Xia et al. (2025) IEEE/CVF; Adriano et al. (2021) ISPRS J. Photogramm. Remote Sens.;

Topic 05

Tsunami Science & Numerical Simulation

Here, we conduct high-resolution numerical simulations of tsunami generation, propagation, and coastal inundation to understand the relationship between tsunami hydrodynamic features and coastal damage. We also study historical events using source inversion based on observed data, such as tide-gauge records and DART buoy data.

Source Inversion Tsunami Simulation Inundation Modeling Field Surveys Fragility Functions

Representative papers: Adriano et al. (2025) Physics Earth Planet. Int.; Mizutani et al. (2025) Geophys. J. Int.; Adriano et al. (2017) Pure Appl. Geophys.; Adriano et al. (2016) Coastal Engineering Journal.