Authors: Sai Harsha Vemula, Arpita Patel, Benjamin Lee, Kamal, Nia, Manjila – Alabama Water Institute, The University of Alabama
Title: Portable Reproducibility for NextGen Water Modeling Using Containerized HPC and Cloud Workflows
Presentation Type: Poster Presentation
Abstract: Reproducibility in High-Performance Computing (HPC) is a cornerstone of robust scientific progress, yet it remains a significant challenge due to complex software stacks, hardware heterogeneity, and intricate dependency management. In the hydrological domain, the Next Generation Water Resources Modeling Framework (NextGen) provides a flexible, model-agnostic engine for continental-scale water prediction. However, deploying NextGen from source requires resolving an extensive dependency graph, making bare-metal reproducibility nearly impossible across diverse HPC environments. To address this, we present NextGen In A Box (NGIAB): a containerized distribution of the NextGen framework utilizing docker and Singularity/Apptainer containers for reproducibility. This paper details the reproducibility of NextGen framework tested in different HPC and cloud infrastructure. NGIAB facilitates sustainable, portable, and exact reproducibility of complex hydrological experiments. We evaluate the framework across five heterogeneous cloud and HPC systems using publicly available GitHub repositories to confirm consistent outcomes. While architectural variations preclude bit-wise identity, we propose a robust numerical acceptance threshold demonstrating that floating-point deviations remain consistent to six decimal places. Our findings confirm that this approach enables transparent, verifiable scientific computing with negligible overhead, significantly lowering the barrier to reliable, large-scale computational research.