CAREER: Development of Geostatistical Data Assimilation Tools for Water Quality Monitoring

Michalak, Anna m .密歇根大学安阿伯分校职业：开发用于水质监测的地质统计数据同化工具在这个时代，任何人都可以在线查看天气预报，以知道他们是否应该为即将到来的周末计划一次野餐，而不会有下雨的风险，为什么不可能登录查看当地海滩的水是否预计在同一天没有大肠杆菌？发展水质预报系统对水资源的长期可持续管理至关重要。为了实现这一目标，需要新的工具以统计严谨的方式合并水质数据，同时最佳地利用现有测量提供的信息。与天气监测和预报不同，由于数据收集的困难和费用，水质评估将始终受到数据相对稀疏的影响。因此，概率框架对于任何水质预测框架的成功都是必不可少的，因为在整个分析过程中需要考虑到与采样水质相关参数相关的不确定性的影响。阻碍概率水质预测框架实施的一个重要知识缺口是缺乏在水质监测网络中吸收不同类型数据的方法。如果能够获得与水质有关的参数分布的数据驱动的统计描述，那么这些信息一旦与水流、输送以及化学和生物相互作用的数值模型相结合，就可以形成水质预测系统的基础。将空间数据同化为数值模型带来了许多自然属于地统计学领域的统计问题。该项目的主要研究目标是通过克服与分析相关的基本限制，如物理约束、支持和缩放问题、不确定性评估和计算问题，开发所需的统计和数值工具，以优化利用稀疏和不完善的水质监测数据。这些研究目标与本项目的教育计划和更广泛的影响密切相关，其中心是向多学科受众广泛传播研究成果，开发创新的教育材料，并强调在科学和工程领域征聘和保留妇女。智力优势：该项目的研究目标集中在新颖的统计严谨工具上，通过创新地使用辅助信息，优化利用有限的水质监测数据。我们的具体特点，典型的水质数据将被解决。(1)开发地质统计马尔可夫链蒙特卡罗地质统计工具，以纳入已知的物理约束并评估其对水质参数分布的影响。(2)可用数据通常具有不同的物理尺度，使得数据集即使测量相同的量也彼此不兼容。该项目将开发适用于水质和相关数据的地统计学降尺度工具。(3)针对海量、多类型、多来源的水质数据，构建卡尔曼滤波平滑统计框架，对水质参数分布进行序次更新估计。(4)将根据第二个和第三个目标的结果，开发用于合并多个数据流的工具。现场数据将用于测试和验证作为前四个目标的一部分开发的单个工具。(5)在项目的最后阶段，开发的统计工具将同时应用于一项试验性实地研究。

Michalak, Anna M.University of Michigan Ann ArborProposal Number: 0644648CAREER: Development of Geostatistical Data Assimilation Tools for Water Quality MonitoringIn a time when anyone can check weather forecasts online to know whether they should plan a picnic for the upcoming weekend without the risk of rain, why is it not possible to log on to check whether the water at the local beach is expected to be free from e-coli on that same day? The development of water quality forecasting systems is essential to long-term sustainable water resource management. In anticipation of this goal, new tools are needed to merge water quality data in statistically rigorous manner while making optimal use of the information provided by the available measurements. Unlike weather monitoring and forecasting, water quality assessment will always suffer from a relative sparsity of data due to the difficulty and expense associated with data collection. As a result, a probabilistic framework is essential to the success of any water quality prediction framework, because the impact of the uncertainty associated with sampled water-quality related parameters needs to be taken into account throughout the analysis.A significant gap in knowledge preventing the implementation of a probabilistic water quality forecasting framework is the lack of methods for assimilating the disparate types of data in a water quality monitoring network. If a data-driven statistical description of the distribution of water-quality-related parameters could be obtained, then this information, once coupled to numerical models of water flow, transport, and chemical and biological interactions, could form the basis of a water quality forecasting system. The assimilation of spatial data into numerical models brings about a number of statistical problems that fall naturally into the realm of geostatistics. The main research goal of this project is to develop the statistical and numerical tools needed to make optimal use of sparse and imperfect water quality monitoring data, by overcoming basic limitations associated with their analysis, such as physical constraints, support and scaling issues, uncertainty assessment, and computational issues. These research goals are closely connected with the educational plan and broader impacts of this project, which center on the broad dissemination of research results to a multidisciplinary audience, the development of innovative educational materials, and the strong emphasis on the recruitment and retention of women in science and engineering.Intellectual merit: The research objectives of this project center on novel statistically rigorous tools formaking optimal use of limited water quality monitoring data, through innovative use of auxiliary information. our specific features typical of water quality data will be addressed. (1) Geostatistical Markov chain Monte Carlo geostatistical tools will be developed for incorporating known physical constraints and assessing their impact on water quality parameter distributions. (2) Available data often have different physical scales, making datasets incompatible with one another even if they are measuring the same quantity. This project will develop tools for geostatistical downscaling applicable to water quality and related data. (3) To deal with large volumes, types and sources of water quality data, a Kalman filtering and smoothing statistical framework will be built for sequentially updating estimates of water quality parameter distributions. (4) Tools for merging multiple data streams will be developed, building on results from the second and third objectives. Field data will be used to test and validate the individual tools developed as part of these first four objectives. (5) In the last phase of the project, the developed statistical tools will be applied concurrently to a pilot field study.