Large-scale hydrological models are demanding both in term of memory allocation and CPU time, particularly when assessment of modeling uncertainty is required. High Performance Computing offers the opportunity to reach resolutions not achievable with standard serial coding. However, the advantages may be offset by poor scalability of the model due to components that have to be executed in series, such as to simulate the presence of hydraulic infrastructures.

Driven by this motivation, we developed HYPERstreamHS, a model that adopts a holistic approach to simulate hydrological processes in large river basins with streamflow altered by hydraulic infrastructures. The model adopts a dual-layer parallelization strategy, where the paralleled version of the hydrological kernel is the first-layer, with the second layer taking care of inverse modeling.

The results show that the processors should be carefully organized and grouped in order to achieve the best overall performance and suggests that this subdivision is problem specific.

大规模水文模型在内存分配和CPU时间方面都要求很高，尤其是在需要评估模型不确定性时。高性能计算提供了达到标准串行编码无法达到的分辨率的机会。但是，由于必须串联执行的组件（例如模拟液压基础设施的存在）而导致的模型可伸缩性差，这些优点可能会被抵消。

在这种动机的驱使下，我们开发了HYPERstreamHS，该模型采用整体方法来模拟大型河流流域的水文过程，并通过水力基础设施改变了水流。该模型采用双层并行化策略，其中水文内核的并行版本为第一层，第二层为逆模型。

结果表明，应该对处理器进行仔细的组织和分组，以便获得最佳的总体性能，并建议此细分是针对特定问题的。