Harnessing high-performance computing and job scheduling for improving SWAT calibration and uncertainty analysis

The Soil and Water Assessment Tool (SWAT) is a conceptual catchment model for hydrological and water-quality simulation that enables relatively fast model runs. However, its large number of parameters representing hydrological and soil processes makes calibration and uncertainty analysis computationally demanding, especially for complex catchments. In this study, we propose a parallel simulation approach for SWAT that integrates parallelization and job scheduling. SWAT-Parallel partitions large tasks into smaller jobs, managed by R-based tools, including SWATrunR and R-SWAT. We evaluated SWAT-Parallel using two catchments of differing complexity. Results indicate that job scheduling, combined with parallelization, significantly reduces model calibration time, with performance constrained primarily by the available resources of the High-Performance Computing (HPC) facility. Beyond a certain computational threshold, further speedup was limited by overhead processes, such as handling input and output data. These findings suggest that optimizing input data management, alongside increasing computational capacity, could further enhance model calibration efficiency.