Methods in machine learning for probabilistic modelling of environment, with applications in meteorology and geology

天气预报的重要性怎么强调都不为过：它对一系列决策都有影响，从带伞到是否应该取消航班或疏散地区。它形成了一个重要的研究领域，汇集了数学，统计学，数据科学和物理学等学科。准确可靠的天气预报是有价值的，但实际上它们依赖于不完善的信息。基于物理学的数学预测模型进行近似，以预测天气，而过去和当前的天气数据无法准确测量，如果有的话。此外，还有许多相互竞争的预测模型（系统），使这项任务更具挑战性。该博士生将专注于调查和开发统计和数据科学技术，以最佳方式混合所有可用于预测天气的信息来源，同时量化所有不确定性和误差来源。该项目将研究一系列方法，以实现最佳合并各种数据源的目标。这些包括（但不限于）统计建模技术，如分层贝叶斯建模，贝叶斯融合和机器学习算法。业务天气预报的另一个挑战是需要任何这样的技术是实用和计算效率。因此，大部分的努力将是在计算机上开发的技术的最佳实施。重点还将放在确定适当的衡量标准上，以便将所开发的技术相互比较，并与目前的“基线”方法进行比较。该项目为学生提供了一个独特的机会，使他们能够在数据科学，统计学和机器学习方面获得高级可转移技能的经验。与此同时，这项工作的动机是提高天气预报准确性的真实的世界挑战，这提供了为英国气象局（UKMO）工作并与之合作的机会。这名学生将在埃克塞特斯特里瑟姆大学校园工作，但也将在埃克塞特的英国气象局工作，与那里的科学家合作。UKMO将提供案例支持，使学生将受益于UKMO天气预报团队的专业知识和天气预报方面的培训。学生还将通过参加英国，欧洲以及国际上的相关研讨会和会议，向研究界传播他们的工作。有一个强大的最终用户元素激励这项工作，特别是可再生能源行业，所以学生也将花时间与天气预报用户，如国家电网谁大力支持这个项目。这个资助的博士奖学金的理想候选人应该具有定量背景，并对数据分析和统计，数据科学和机器学习等领域感兴趣。他们还应该对计算感兴趣，因为据设想，所开发的技术将在最先进的云计算上实施。该奖学金包括英国学费以及每年14，553英镑的生活津贴。它还包括培训课程等发展和旅行（如参加会议）的保险。

The importance of weather forecasting cannot be overstated: it has impacts on a range of decision making, from taking an umbrella to whether flights should be cancelled or areas evacuated. It forms an important area of research which brings together disciplines such as mathematics, statistics, data science and physics. Accurate and reliable weather forecasts are valuable, but in practice they rely on imperfect information. Mathematical forecasting models based on physics make approximations in order to forecast weather, while past and current weather data cannot be measured accurately, if at all. In addition, there many competing forecasting models (systems) making the task even more challenging. This PhD studentship will focus on investigating and developing statistical and data science techniques to optimally blend all sources of information available for forecasting weather, while at the same time quantifying all sources of uncertainty and error. The project will look at a range of methods for achieving the goal of optimally merging various data sources. These include (but are not limited to) statistical modelling techniques such as hierarchical Bayesian modelling, Bayesian melding and machine learning algorithms. An added challenge to operational weather forecasting is the need for any such technique to be practical and computationally efficient. Therefore much of the effort will be on the optimal implementation of the developed techniques on the computer. Emphasis will also be on defining suitable metrics with which to compare the developed techniques with each other as well are with current "baseline" approaches. This project offers a unique opportunity for a student to gain experience in advanced transferable skills across data science, statistics and machine learning. At the same time, the work is motivated by the real world challenge of improving weather forecast accuracy, which provides the opportunity to gain experience in working for and with the UK Met Office (UKMO). The student will be based at the University of Exeter Streatham campus, but will also be expected to spend time at the UKMO in Exeter, collaborating with scientists there. The UKMO will provide CASE support so that the student will benefit from expertise in the weather forecasting team at the UKMO and training with respect to weather forecasting. The student will also be expected to disseminate their work to the research community by attending relevant workshops and conferences across the UK, Europe but also internationally. There is a strong end-user element motivating this work, specifically the renewable energy industry, so the student will also be spending time with weather forecast users such as National Grid who strongly support this project. The ideal candidate for this funded PhD scholarship should have a quantitative background and be interested in data analysis and fields such as statistics, data science and machine learning. They should also have an interest in computing, as it is envisioned that the techniques developed will be implemented on state-of-the-art cloud computing. The scholarship includes UK tuition fees as well as £14,553 maintenance allowance per year. It also includes cover for development such as training courses and for travel (e.g. attending conferences).