Atmospheric modeling and data assimilation

在大气科学中，资料同化的目的是调和大气的测量结果和从我们所知的代表大气动力学的数值模式得到的预报。同化同时考虑了观测和预报的误差统计量。在过去的二十年中，数据同化系统已经发展，现在可以摄取非常大量的数据（每天约107个）并考虑到时间维度。在过去的几年里，我一直在研究一种卫星仪器的原型，用来测量远红外辐射。为了优化仪器的配置以从这些测量中获得最多的信息，基于信息论开发了一种简单的方法来评估仪器的不同配置。初步研究可以确定应该选择哪些波段。考虑到在卫星上放置仪器的高成本，有工具帮助我们做出正确的选择是非常有帮助的。新类型观测的工作将继续进行，目的是将1D方法与更复杂的观测系统实验进行比较，在这种实验中，通过在两个实验之间增加或减少观测值，将现有观测值吸收到最先进的系统中，时间长达数周至数月。目的是评估人们对1D更简单方法的信任程度。本研究的另一个方面是在对照实验中使用更简单的模型来评估不同的数据同化策略的优缺点。重点将放在通过观测发现可能与高影响天气事件有关的迅速发展的扰动的能力。为了能够捕捉如此微弱的信号，需要新的方法，但它们需要大量的资源才能适用于复杂的业务天气预报系统，这些系统现在将大气与其他成分（例如海洋、陆地）耦合在一起。但从理论到应用还有一个巨大的飞跃。在卫星数据同化联合中心（JCSDA）的合作伙伴（例如NASA）之间的合作开发的数据同化集成联合努力（JEDI）中，制定了一项战略，可以对不同的同化方法进行测试，并与简单模型或完整的NWP模型进行比较。目标是在更好的控制环境中开发数据同化，正如我所建议的那样，可以用更复杂的系统进一步测试。科学界认为这是发展数据同化系统的第一步，也是必要的一步。在受控环境中使用简单模型可以帮助我们更好地理解天气系统快速发展背后的动力学，或者理解气候为何有时会缓慢而突然地演变。这就是这项研究的首要目标。

The objective of data assimilation in atmospheric science is to reconcile measurements of the atmosphere and a forecast from a numerical model which represents the dynamics of the atmosphere to the best of our knowledge. The assimilation takes into account the error statistics of both the observations and the forecast. Over the last two decades, data assimilation systems have evolved and can now ingest very large volumes of data (~107 per day) and take into account the time dimension. During the last few years, I have been working on a prototype satellite instrument to measure radiances in the far infrared. To optimize the configuration of the instrument to obtain the most information from those measurements, a simple approach has been developed based on information theory to evaluate different configurations of the instrument. A preliminary study allowed to determine which wavebands should be selected. Given the high cost of putting an instrument on a satellite, it is very helpful to have the tools to help us make the right choices. The work on new types of observations will be pursued with a view to compare the 1D approach to more elaborate Observing System Experiments  in which existing observations are assimilated in state-of-the-art systems by adding or subtracting observations between two experiments over periods of weeks to months. The objective will be to assess the degree of confidence one may have in the 1D simpler approach. Another aspect of the proposed research is concerned with different data assimilation strategies in controlled experiments with simpler models to evaluate their strengths and weaknesses. Emphasis will be put on the ability to detect through observations the emergence of rapidly developing disturbances that can be associated with high impact weather events. To be able to capture such a weak signal, new methods are needed but they require  a significant amount of resources to be applicable to complex operational weather forecast systems which now couples the atmosphere with other components (e.g., oceans, land). But there is a big leap from theory to applications. In the Joint Effort for Data assimilation Integration (JEDI), a collaborative development between partners (e.g., NASA) of the Joint Center for Satellite Data Assimilation (JCSDA), a strategy is developed in which different assimilation methods can be tested and compared with either a simple model or a complete NWP model. The objective is for data assimilation developed in better controlled environments, as the one I propose, can be further tested with more complex systems. The scientific community sees this as a first and necessary step in the development of data assimilation systems. Using simple models in a controlled environment can help to better understand the dynamics underlying the rapid development of a weather system or why climate evolves slowly but abruptly at times. This is what the overarching objective of this research is about.