Lay Summary

Computer models predicting complex phenomena—like climate impacts, air pollution, or disease spread—are inherently subject to uncertainties in their input data. Understanding how these uncertainties affect predictions is crucial for policy decisions. This paper introduces the mUQSA toolkit, a user-friendly software framework designed for Uncertainty Quantification (UQ) and Sensitivity Analysis (SA) on high-performance computers. By automating the execution of thousands of simulation variations, mUQSA helps researchers identify which parameters most strongly influence outcomes, yielding far more reliable predictions for global challenge modeling. (84 words)

Full Abstract

Predictive simulations addressing global challenges suffer from inherent uncertainties stemming from parameter approximation, noisy input data, and model abstractions. Uncertainty Quantification (UQ) and Sensitivity Analysis (SA) are indispensable for evaluating model reliability, but their computational cost on HPC systems is often prohibitive. We introduce mUQSA (micro Uncertainty Quantification and Sensitivity Analysis), a lightweight, scalable toolkit integrated into HPC workflows. mUQSA automates sample generation, distributed execution, and statistical post-processing. We demonstrate its effectiveness across multiple environmental and fluid dynamics benchmarks, showcasing how automated UQ empowers policy-makers with probabilistic insight.

Metadata

• Publication Date: 2024
• Author:   Michał Kulczewski, Bartosz Bosak, Piotr Kopta, Wojciech Szeliga, Tomasz Piontek
• Journal: Journal of Computational Science