High performance computing framework for GCM-driven climate change simulation with the routing model WATROUTE

马尼托巴省水电公司的资源规划过程受到各种因素的影响，包括供水和电力需求。因此，了解规划决策对未来水流情景的敏感性对于它们的综合资源规划过程非常重要。通过将全球气候模型模拟的径流值与纳尔逊-丘吉尔流域的WATROUTE路线模型相结合，可以开发径流场景。路径模型在WATROUTE中实现，WATROUTE是加拿大水文模型(CHALM)程序套件的路径组件。此外，WATROUTE是径流模拟过程的计算瓶颈，导致运行时间较长，特别是在预测多年后的未来时。因此，找到显著缩短WATROUTE运行时间的方法对马尼托巴省水电公司来说非常重要。此外，路由模型的并行本质非常适合在图形处理单元(GPU)中固有的大规模并行处理器上执行，这些处理器最近已可用于通用科学计算。这项工作是与Manitoba Hydro的Engage Grant的结果的扩展，该项目的目标是创建WATROUTE路由模型的高度并行、基于GPU的高性能计算实施，以帮助Manitoba Hydro的资源规划过程。通过显著减少路由模型的运行时间，朝着这一目标取得了巨大的进步。因此，此次Engage Plus Grant的目的是追求使用GPU减少WATROUTE的总体运行时间，生产WATROUTE的发布就绪版本，探索将WATROUTE应用于以前太大而无法处理的数据集，并调查WATROUTE在具有多个GPU的系统上的使用情况。这项工作实现的好处也可能对MEC-地表和水文系统(MESH)和加拿大区域气候模式产生积极影响，因为这两个系统都使用WATROUTE。

Manitoba Hydro's resource planning process is influenced by a variety of factors including water supply andelectrical energy demand. As a result, understanding the sensitivity of planning decisions to future streamflowscenarios are important to their integrated resource planning process. Streamflow scenarios can be developedby combining global climate model simulated runoff values with the WATROUTE routing model for theNelson-Churchill Watershed. The routing model is implemented in WATROUTE, the routing component ofthe Canadian Hydrological Model (CHARM) suite of programs. Further, WATROUTE is the computationalbottleneck for the streamflow simulation processes, incurring long runtimes, especially when projecting manyyears into the future. Thus, finding ways of significantly reducing the runtime of WATROUTE is of greatimportance to Manitoba Hydro. Moreover, the parallel nature of the routing model is an ideal fit for executionon the massively parallel processors inherent in graphics processing units (GPUs), which have recently becomeavailable for general-purpose scientific computing. This work is an extension of the results from an EngageGrant with Manitoba Hydro, where the aim was to create a highly-parallel, GPU-based,high-performance-computing implementation of the WATROUTE routing model to assist Manitoba Hydro'sresource planning process. Tremendous progress was achieved toward this aim through significant reduction inthe routing model runtime. As a result, the aim of this Engage Plus Grant is to pursue the reduction ofWATROUTE's overall runtime using GPUs, produce a release ready version of WATROUTE, explore theapplication of WATROUTE to datasets previously too large to process, and investigate the use ofWATROUTE on systems with multiple GPUs. Benefits realized by this work may also have a positive impacton the MEC-Surface and Hydrology System (MESH) and the Canadian Regional Climate Model, as both useWATROUTE.