CMG Research: Making inferences about planktonic ecosystems with models and observations: use of emulators to make complex multidimensional applications tractable

该奖项是根据2009年《美国复苏和再投资法案》(公法111-5)资助的。仿真器是复杂模型输出作为不确定输入函数的计算廉价表示。它们提供了极大的计算节省，因此可能会极大地扩大使用最先进的、空间上明确的浮游生物生态系统模型进行统计推断的可能性--然而它们以前从未在这方面被利用过。浮游生物模型的多维性要求对当前的模拟器方法进行重大扩展。因此，这一领域的研究将是全新的，需要浮游生物生态学家和数学家之间的密切合作。拟议的研究集中在三个具体的应用上，所有这些都超出了现代计算资源的范围，没有使用仿真器或替代的快速逼近。首先，仿真器将被用来测试简单的浮游生物生态系统模型在一维空间环境中适应百慕大大西洋时间序列(BATS)观测的能力。这个问题为开发多维输出的仿真器方法提供了理想的试验台。这将从假设作用力已知的应用开始。在这种情况下，动态模型相对便宜，允许将使用仿真器得出的推论与使用动态模型得出的推论进行比较。然后，仿真器的计算优势将在不完全了解的强迫的背景下得到利用，从而促进对生物过程和物理强迫中的不确定性的联合推断。这将使研究人员能够解决浮游生物生态系统建模中的一个长期存在的问题：需要多大的物理和生物复杂性来解释BATS数据？第二个应用程序旨在开发一个用于整个北大西洋的3D浮游生物模型的模拟器，并使用它在已知物理强迫的假设下从卫星数据推断模型参数值。这将允许推断整个流域浮游生物丰度、初级生产力和碳输出的季节性变化。第三，将结合全流域模型使用仿真方法来评估各种气候变化情景下生态系统反应预测的不确定性。这一努力的学术价值源于对海洋学和统计学之间的交界点问题的跨学科方法。利用先进的统计和数据同化方法将模型与观测结合起来，将产生对浮游生物生态在地方到盆地尺度上的新见解，提供与海洋生态系统和全球碳循环有关的信息。通过这样做，拟议的研究将促进这两个学科及其相互之间的联系。这项活动的广泛影响包括培养研究生和本科生以及博士后研究员。课程发展将包括加强现有的关于将浮游生态系统中的模型和观测联系起来的课程。私人投资司还计划编写一份简短的文本，以促进这一材料的广泛传播。

This award is funded under the American Recovery and Reinvestment Act of 2009 (Public Law 111-5).Emulators are computationally inexpensive representations of complex model output as a function of uncertain inputs. They provide drastic computational savings, and may thus greatly expand the possibilities for statistical inference using state-of-the-art, spatially-explicit plankton ecosystem models - yet they have never previously been utilized in this context. The multidimensional nature of plankton models requires significant extension of current emulator methods. Research in this area will therefore be entirely novel and require a close collaboration between plankton ecologists and mathematicians.The proposed research focuses on three specific applications, all of which are beyond the scope of modern computational resources without the use emulators or alternative fast approximations. First, emulators will be used to test simple plankton ecosystem models in terms of their ability to fit the Bermuda Atlantic Time Series (BATS) observations in a one-dimensional spatial setting. This problem provides an ideal testbed for developing emulator methods for multidimensional output. This will begin with an application in which the forcing is assumed to be known. In this context, the dynamical model is relatively inexpensive, allowing the inferences drawn using the emulator to be compared with those drawn using the dynamical model. The computational advantage of the emulators will then be exploited in the context of forcing that is not perfectly known, facilitating joint inference of biological processes and uncertainty in physical forcing. This will allow the investigators to address a long-standing question in plankton ecosystem modeling: How much physical and biological complexity is necessary to explain the BATS data?The second application aims to develop an emulator for a 3D plankton model of the entire North Atlantic, and use it to infer model parameter values from satellite data under the assumption of known physical forcing. This will permit inference of basin-wide seasonal variability in plankton abundance, primary production, and carbon export.Third, emulation methods will be used together with the basin-wide model to assess uncertainty in forecasts of ecosystem response for various climate change scenarios. The intellectual merit of this effort stems from an interdisciplinary approach to problems at the interface between oceanography and statistics. Conjoining of models with observations using advanced statistical and data assimilative methodologies will result in new insights into plankton ecology on local to basin scales, providing information relevant to oceanic ecosystems and global carbon cycling. In so doing, the proposed research will advance both disciplines as well as their interface.Broader impacts of this activity include training of graduate and undergraduate students as well as a postdoctoral fellow. Curricular development will include enhancement of an existing course on linking models and observations in planktonic ecosystems. The PIs also plan to produce a short text that will promote broad dissemination of this material.