Development of advanced calibration methods for computationally expensive hydrologic simulation models

环境或水文模拟模型是复杂环境系统的数学近似。 健全的环境管理决定，例如如何减轻气候变化对水资源的未来影响，通常是以环境模拟模型预测为基础的。我的研究产生了支持这些模型开发的先进工具。 使用我的工具的高级建模者应该能够更快地构建更好的模型。 更重要的是，我的工作支持这些模型的应用，当选定的模型应用于给定的案例研究时，我的工具确保模型对过去系统行为的预测尽可能准确。 准确预测过去的系统行为往往是使用模拟模型为环境决策提供信息的先决条件。 我们的一些最先进的水文系统模型是如此复杂，他们可能需要几个小时才能在个人电脑上运行。 不幸的是，这种长的运行时间常常妨碍了它们在决策中的使用。 这项研究资助支持继续开发工具，这些工具适用于非常大且计算昂贵的仿真模型。 我的研究将产生多种新的水文模型校准方法，这些方法适用于各种计算规模，并可供研究人员和从业人员使用。 这些新的方法和工具也将考虑模型预测的不确定性，因为我们的数学描述的性质是不完善的，我们的系统行为的测量也是不完善的。 本研究的结果适用于任何环境模型的校准问题，以及其他工程学科与类似的模型校准要求。 这项工作的结果将特别使加拿大受益，因为研究中使用的主要建模案例研究之一是加拿大环境部的五大湖水位和流出量预测和模拟模型。

Environmental or hydrologic simulation models are mathematical approximations of complex environmental systems. Sound environmental management decisions such as how to mitigate future impacts of climate change on water resources are typically based on environmental simulation model predictions. My research generates advanced tools that support the development of these models. Advanced modellers using my tools should be able to build better models faster. Even more important, my work supports the application of these models such that when a selected model is applied to a given case study, my tools make sure the model predictions of the past system behaviour are as accurate as possible. Accurate predictions of past system behaviour is often a prerequisite for using a simulation model to inform environmental decision-making. Some of our most advanced models of hydrologic systems are so complex they can require hours to run on a personal computer. Unfortunately, this long run time often precludes their use in decision-making. This research grant supports the continued development of tools that are tailored for application to very large and computationally expensive simulation models. My research will generate multiple new hydrologic model calibration approaches that are adaptable to a wide range of computational scales and made accessible to both researchers and practitioners. These new approaches and tools will also be developed considering model prediction uncertainty since our mathematical descriptions of nature are imperfect and our measurements of system behaviour are also imperfect. Results of this research are applicable to any environmental model calibration problem as well as other Engineering disciplines with similar model calibration requirements. Results of this work will specifically benefit Canada since one of the major modelling case studies utilized in the research is Environment Canada's forecasting and simulation model of Great Lakes levels and outflows.