High-fidelity hydrological models are increasingly built and used to investigate the effects of management activities and climate change on water availability and quality for large areas with datasets of high spatial and temporal resolution. However, these advantages come at the price of greater computational demand and run time. This becomes challenging when modeling routines involve iterative model simulations. In this study, we proposed a generic scheme to reduce the Soil and Water Assessment Tool (SWAT) runtime by decomposing a watershed model into subbasin models and optimizing the subbasin model simulations based on a parallel approach. Based on this scheme, we implemented a generic tool named Spark-SWAT, which allows subbasin models to be simulated in parallel on a Spark computer cluster. We then evaluated Spark-SWAT with two sets of experiments to demonstrate the potential of Spark-SWAT to accelerate single and iterative model simulations. In each test set, Spark-SWAT was applied to simulate 12 synthetic hydrological models in parallel with different I/O (input/output) burdens and river network complexities in a Spark cluster with five virtual machines. The single model parallelization results showed that Spark-SWAT yielded a speedup value of 7.84 for the most complex model but was less effective with simple models. When applied to use cases with iterative model runs, Spark-SWAT yielded a speedup of 6.55–24.58 depending on the model complexity. These results indicate that the proposed scheme can effectively solve high-computational-demand problems of complex models. As a subbasin-level parallelization tool, Spark-SWAT can be very computationally frugal and useful in use cases in which the model input changes pertain to only a few subbasins because only the changed and downstream subbasins require new computations. Moreover, it is possible to apply this generic method to other subbasin-based hydrological models to alleviate I/O demands and optimize model computational performance.

越来越多地建立和使用高保真水文模型，以研究管理活动和气候变化对具有高空间和时间分辨率数据集的大面积地区水资源可用性和质量的影响。然而，这些优势是以更大的计算需求和运行时间为代价的。当建模例程涉及迭代模型模拟时，这变得具有挑战性。在这项研究中，我们提出了一种通用方案，通过将流域模型分解为子流域模型并基于并行方法优化子流域模型模拟来减少土壤和水评估工具 (SWAT) 运行时间。基于这个方案，我们实现了一个名为 Spark-SWAT 的通用工具，它允许在 Spark 计算机集群上并行模拟子流域模型。然后，我们通过两组实验评估了 Spark-SWAT，以证明 Spark-SWAT 在加速单一和迭代模型模拟方面的潜力。在每个测试集中，Spark-SWAT 被应用于在具有五个虚拟机的 Spark 集群中并行模拟 12 个合成水文模型，这些模型具有不同的 I/O（输入/输出）负担和河网复杂性。单模型并行化结果表明，对于最复杂的模型，Spark-SWAT 产生了 7.84 的加速值，但对简单模型的效果较差。当应用于具有迭代模型运行的用例时，Spark-SWAT 根据模型复杂性产生了 6.55-24.58 的加速。这些结果表明，所提出的方案可以有效地解决复杂模型的高计算需求问题。作为子流域级并行化工具，Spark-SWAT 在计算上非常节俭，并且在模型输入更改仅涉及少数子流域的用例中非常有用，因为只有更改的子流域和下游子流域需要新的计算。此外，可以将此通用方法应用于其他基于子流域的水文模型，以减轻 I/O 需求并优化模型计算性能。