HiDALGO2 D3.8 Ensemble Scenarios for Global Challenges

The deliverable, D3.8, presents the final consolidated report on ensemble scenario developments within the HiDALGO2 project. Building on the foundations established in D3.7, the document covers the methodological framework, enabling technologies, and pilot-specific implementations of ensemble-based simulation across five application domains. A shared ensemble lifecycle - define, generate, execute, analyse, use — is identified across the contributing pilots, while the specific methods and tools at each stage reflect the diversity of the underlying physical systems and decision-making contexts. For each pilot, specific ensemble implementations have been developed and demonstrated:

• Urban Air Project (SZE): A full-factorial parametric ensemble of 48 steady-state NO₂ dispersion simulations has been executed over a 10.6 million cell mesh of the Győr city geometry, varying 16 wind directions and 3 wind speeds. Singular Value Decomposition of the resulting concentration fields identifies the dominant spatial dispersion patterns and the effective degrees of freedom, providing a basis for reduced-order modelling and scenario-based urban air-quality assessment.
• Urban Building (UNISTRA): A reproducible ensemble framework has been integrated into the KUB simulation engine, supporting Monte Carlo and Latin Hypercube sampling with variance reduction, MPI-based parallel execution, and online statistical reduction using the Welford/Chan algorithm. The framework provides city planners with robust building rankings, exceedance probabilities, and quantile-based uncertainty indicators for energy demand and thermal comfort assessment.
• Renewable Energy Sources (PSNC): Sensitivity analyses with up to 625 ensemble members have been conducted on up to 70,000 CPUs using the mUQSA toolkit and QCG-PilotJob orchestrator. Through Polynomial Chaos Expansion and Sobol indices, the analysis demonstrates that wind direction is the dominant uncertainty driver for local wind-speed predictions, while temperature and pressure perturbations have negligible influence.
• Wildfires (MTG): The 50-member ECMWF MARS meteorological ensemble has been integrated with multiple ignition-point scenarios and planetary boundary layer parameterization experiments, producing the first comprehensive probabilistic assessment of wildfire behaviour under combined meteorological, ignition, and model-physics uncertainties.
• Material Transport in Water (FAU): A systematic parameter-sweep approach has been implemented using the waLBerla framework, comprising 96 Couette flow simulations for model validation and over 100 thermally stratified simulations for AI surrogate data generation, executed across CPU and GPU architectures on EuroHPC systems.

Taken together, the work across the five pilots demonstrates that ensemble-based uncertainty quantification is computationally feasible at HPC scale, and that the resulting outputs provide a solid foundation for the downstream AI, HPDA, and uncertainty quantification activities in Work Package 4.