Assistant Professor
VESCOVO Ruben
Organization
Research Division
- Risk Evaluation and Disaster Mitigation Research Division
- Disaster Humanities and Social Science Division
- Disaster Medical Science Division
- Practical Research and Collaboration Division
- Endowed Research Division
- Joint Research Division
- Co-creation Center for Disaster Resilience
- Global Centre for Disaster Statistics
- Kesennuma Satellite Office
- Public Relations Office
- Administration Office
- Research Subject(s)
- Complex systems applied to evaciation modelling. Multiagent based simulations, decision making under uncertainty, computational social psychology.
- Key Words
- Agent Based Modelling / Evacuation / Bayesian statistics.
- Website
- Research Activities
-
Multi-agent simulation for evacuation logistics
I work in the intersection between social psychology, computer science, and engineering. Our main goal is to reliably simulate disaster driven evacuation & response, accounting for realistic social dynamics of the population, organizational capacities of responders, and the capabilities of current institutional frameworks and logistics.
As such, we aim to better understand limitations, identify bottlenecks, and derive potential improvements to current frameworks and systems.
Physically informed probabilistic tsunami damage estimation
I leverage probabilistic machine learning techniques informed by Bayesian methods of optimal decision making under uncertainty to estimate potential damage given tsunami drivers and environmental conditions (inundation, ground elevation, land use, building configuration, etc.).
The probabilistic approach allows us to identify uncertainty and credible bands over our prediction, which we can leverage to make optimal decisions under a most likely scenario.
Beyond the immediate estimate, we can again leverage the uncertainty to further investigate latent processes and iterate over our process, to improve its predictive accuracy.
Social sensing in the post disaster window
I leverage international news media reports and large language models to automatically mine information of affected buildings, locations, and number of affected people. This information is compressed and processed using natural language processing, to extract named entities, aggregate mentions through lexical and semantic similarity, and finally geolocate buildings and affected areas. This method is fully automated and resiliant even under severe information blackouts.
- Selected Works
-
Vescovo, R., Adriano, B., Wiguna, S., Ho, C. Y., Morales, J., Dong, X., Ishii, S., Wako, K., Ezaki, Y., Mizutani, A., Mas, E., Tanaka, S., and Koshimura, S.: The 2024 Noto Peninsula earthquake building damage dataset: Multi-source visual assessment, Earth Syst. Sci. Data Discuss. [preprint], https://doi.org/10.5194/essd-2024-363, in review, 2025.
Vescovo, R., Mas, E., Adriano, B., Koshimura, S. (2023): Deep learning of tsunami building damage from multimodal physical parameters for real-time damage assessment, XXVIII General Assembly of the International Union of Geodesy and Geophysics (IUGG) (Berlin 2023).
https://doi.org/10.57757/IUGG23-2256Vescovo, R., Adriano, B., Mas, E. et al. Beyond tsunami fragility functions: experimental assessment for building damage estimation. Sci Rep 13, 14337 (2023). https://doi.org/10.1038/s41598-023-41047-y
S. Wiguna, B. Adriano, E. Mas and S. Koshimura, "Evaluation of deep learning models for building damage mapping in emergency response settings", IEEE J. Sel. Topics Appl. Earth Observ. Remote Sens., vol. 17, pp. 5651-5667, 2024.
- Selected Memberships
-
- European Geosciences Union
- American Geophysical Union
- Japan Society of Civil Engineers
- Engineering Australia