A shift to genuinely multi-hazard, multi-risk disaster management is both practicable and worthwhile, but meaningful uptake needs coordinated progress on methods, usable tools, equity, local coproduction and early-career capacity. Without standardized concepts, stronger spatio-temporal evidence, and operationally tested decision-support systems, benefits will remain patchy and context-dependent.

Main Finding

MYRIAD‑EU (2021–2025) demonstrates that moving from siloed single‑hazard approaches to a co‑produced, flexible, systems‑level multi‑hazard and multi‑risk approach is feasible and useful for decision making, but requires (a) common, living terminology and ontologies, (b) governance and co‑production processes, (c) tailored methods and tools that combine physical and social dynamics, and (d) stronger emphasis on usability, equity, and continual testing in real case studies.

Key Points

• Research agenda and progress
  • The project operationalized six research priorities from Ward et al. (2022): common definitions; co‑developed multi‑risk framework; dynamics of hazard/exposure/vulnerability; future multi‑hazard scenarios; cross‑hazard assessment of DRM measures; testing in in‑depth pilots.
• Terminology and knowledge sharing
  • Produced a co‑developed Handbook/glossary (140 terms) and emphasised that definitions must remain “living” to reflect practice and sectoral specifics.
  • Linked glossary/tools into the Disaster Risk Gateway wiki for community updates and resource sharing.
• Framework and co‑production
  • Co‑developed and iteratively tested a flexible six‑step system-of-systems framework for single → multi → systemic risk; implemented via participatory protocols and an online dashboard.
  • Framework supports integration of direct and indirect (cascading) risks and helps identify inter‑system gaps and actor roles.
• Methods & tech advances
  • Assembled a Metadatabase for Dynamics of Risk Drivers cataloguing methods and datasets developed in MYRIAD‑EU: multi‑hazard dynamic footprints, time‑dependent intensity–damage functions, vulnerability indicators, exposure dynamics (time‑since‑previous event).
  • Demonstrated use of ML (e.g., DBSCAN clustering) for spatiotemporal multi‑hazard footprints; developed Dynamic Adaptive Policy Pathways – Multi‑Risk (DAPP‑MR) for policy planning.
• Pilots and governance
  • Five Pilots (North Sea, Canary Islands, Scandinavia, Danube, Veneto) co‑produced methods with stakeholders across sectors (energy, infrastructure, tourism, ecosystems, finance, food/agriculture).
  • Governance review highlighted that National Risk Assessments often remain siloed; platforms exist but cross‑sectoral governance is fragmented.
• Usability and equity gaps
  • Many tools remain technically sophisticated and early‑stage; practitioner usability and contextual relevance are limiting factors.
  • Explicit inclusion of equity and distributive outcomes in DRM remains insufficient and must be mainstreamed.
• Practical outputs and uptake
  • Online dashboard and pilot examples; parts of the framework and terminology are already adopted by other EU projects; emphasis on Multi‑Hazard Early Warning Systems and continued case testing.

Data & Methods

• Co‑production methods
  • Participatory workshops, stakeholder interviews, iterative testing in five regional pilots, sectoral engagement to co‑design terms, indicators, and governance insights.
• Reviews and synthesis
  • Comprehensive literature and tool reviews; policy and governance analyses across European contexts.
• Framework & operational tools
  • Six‑step multi‑risk framework with guidance protocols; online dashboard for navigation and examples; Disaster Risk Gateway wiki for living content.
• Empirical & computational methods assembled in the Metadatabase
  • Multi‑hazard dynamic footprints and susceptibility maps (example: unsupervised clustering — DBSCAN for spatiotemporal clustering of events).
  • Time‑dependent intensity–damage functions (capturing changing vulnerability of assets over time).
  • Vulnerability indicators (quantitative and qualitative) and exposure as function of time since previous hazard (path dependence).
  • DAPP‑MR: extending Dynamic Adaptive Policy Pathways to multi‑risk contexts to identify adaptation pathways under deep uncertainty.
• Pilots as testbeds
  • Contextualized datasets (hazard records, exposure maps, sectoral dependencies), stakeholder feedback loops, and scenario testing across temporal horizons.
• Limitations noted
  • Tool maturity and practitioner usability; language/translation barriers (Handbook primarily in English); need for standardized, interoperable ontologies; persistent data gaps and contested definitions in application.

Implications for AI Economics

How economists and AI researchers can leverage MYRIAD‑EU findings and where AI‑economics research is needed:

1. Integrating multi‑hazard scenarios into economic risk models

   • Use MYRIAD‑EU scenario outputs and metadatabase components as inputs to macro and sectoral stress tests (finance, insurance, supply chains, tourism).
   • Develop stochastic scenario generators that produce correlated, cascading hazard shocks to test systemic economic vulnerability.

2. ML for spatiotemporal hazard–exposure–vulnerability dynamics

   • Extend unsupervised methods (e.g., DBSCAN) and temporal ML to detect event clusters and evolving footprints; combine with spatiotemporal econometric models to estimate economic exposure dynamics.
   • Build hybrid physics‑ML models (scientific priors + data‑driven components) for hazard intensities and time‑dependent damage functions.

3. Modeling feedbacks and systemic contagion

   • Map inter‑sectoral dependencies (from Pilots) into network/agent‑based economic models to study cascading failures (e.g., infrastructure → energy → production).
   • Adapt financial systemic risk methods (contagion, co‑value‑at‑risk) to cross‑sector, multi‑hazard contexts.

4. Decision support under deep uncertainty

   • Translate DAPP‑MR and robust decision frameworks into algorithmic tools: robust optimization, multi‑objective reinforcement learning with human‑in‑the‑loop constraints, and adaptive policy algorithms that produce policy pathways rather than single optimal policies.
   • Emphasise explainability and narrative outputs to match co‑production needs.

5. Usability, explainability, and co‑production in model design

   • Prioritise interpretable models and user interfaces (dashboards, scenario walkers) co‑designed with policymakers and practitioners to improve uptake.
   • Incorporate human‑centered evaluation metrics (ease of communication, decision relevance).

6. Equity, distributional impacts and fairness

   • Embed distributional metrics into economic impact models (who bears losses; long‑term recovery differences).
   • Develop algorithmic fairness assessments for adaptation and allocation policies; enable scenario analyses that quantify unequal outcomes across socio‑economic groups.

7. Data, standards, and benchmarking

   • Use MYRIAD‑EU’s living glossary and Disaster Risk Gateway as a foundation for common ontologies and benchmark datasets for multi‑hazard economic modeling.
   • Create open, standardized datasets that combine hazard footprints, local exposure, asset vulnerability, and socio‑economic indicators to enable reproducible AI‑economics research.
   • Address non‑stationarity (climate change) via continual learning and domain‑adaptation techniques; quantify uncertainty from distributional shift.

8. Applications for finance and insurance

   • Design multi‑hazard portfolio risk models and capital stress tests; quantify tail dependence across hazards and sectors.
   • Evaluate cross‑policy synergies and maladaptation risks (e.g., measures that reduce one hazard exposure but increase another).

9. Research agenda / practical tasks for AI economists

   • Build hybrid models for time‑dependent damage functions with uncertainty quantification.
   • Develop scenario generators for cascading multi‑hazard shocks suitable for macro stress tests.
   • Construct interoperable toolchains (ontologies + APIs) that allow policymakers to plug in regional data and get interpretable risk pathways.
   • Pilot co‑development with local stakeholders to validate models and ensure practical relevance (follow MYRIAD Pilot methodology).
   • Integrate equity metrics into decision‑making algorithms (cost–benefit analyses that include distributional weights).

Overall, MYRIAD‑EU provides a practical blueprint and curated resources (handbook, metadatabase, dashboard, pilots) that AI‑economics researchers can use to build models and decision tools that explicitly handle correlated, dynamic, and cascading hazards — but success requires attention to co‑production, interpretability, data standards, and equity.