Change in the Atlantic Atmosphere Ocean System: ChAAOS

该项目使用两种新的投入，以解决在全球大气-海洋耦合系统的背景下理解大西洋观测到的可变性和变化这一根本问题。利用分布式计算实现的大型整体扰动物理实验和对最近海洋状态的新的基于伴随的估计以及其中的不确定性，我们的目标是识别自由运行的大气-海洋环流模型版本，该模型首次实际再现了过去15年来大尺度气候系统的全部演变。这些模式的集合将为最近海洋状况变化的起源、性质和可预测性提供独特的见解，并为评估未来的可预测性和较长期的大西洋经向翻转环流变化的风险提供宝贵的工具。目前用于十年预测和对不断变化的边界条件的响应的较长期预测的耦合模式通常依赖于模式气候学的模式异常与通过对真实世界的气候学的某种估计而观察到的异常的比较。在以下情况下，这是一个根本问题：(A)对外部强迫的反应是不确定的，与初始状态可能产生的任何可预测性相当，或者(B)系统包含可能对任何预测产生影响的重大非线性。我们将使用一种全新的方法来直接从最先进的海洋分析中初始化耦合模型，使用耦合模型参数的直接扰动来找到跟踪过去15年真实世界的模型版本。这一极具挑战性的目标是由于前所未有的计算资源而成为可能的，允许使用分布式计算提供的完全耦合的大气-海洋大气环流模式来进行数千人的集合。我们的方法将是初始化来自ECCOc海洋分析的两个AOGCM的数万到数十万个扰动版本，考虑到观测和结构的不确定性，从ECCOc和其他分析之间的差异估计。我们将使用似然分析和重要性抽样的统计技术来确定参数/分析组合，这些组合允许模型在最初的六个月内继续‘跟踪’分析，随后，当我们接近有希望的扰动时，一直持续到整整15年。使用的模型将是HadCM3，它已经为分布式计算应用程序建立，以及一个新的模型，该模型基于HadAM3P模型与ECCOc分析中使用的MITgcm的耦合，利用了从过去的系综实验中已经获得的关于两个模型的参数敏感性的信息，以及(在MITgcm的情况下)从模型伴随项获得的信息。然后，成功的模式版本将在整个20世纪(HadCM3)或从1975年起(HadAM3P/MITgcm)免费运行，以评估它们仅在过去20年响应总的外部强迫和人为强迫而产生的AMOC趋势的范围。它们还将为Valor财团将要进行的集合预报实验提供一系列初始条件。除了科学上的好处，该项目还将提供一个重要的公共宣传机会，使参与者能够看到直接使用快速数据来解决明确和紧迫关切的问题。

This project uses two novel inputs to address the fundamental problem of understanding observed variability and change in the Atlantic Ocean in the context of the global coupled atmosphere-ocean system. Using a combination of large-ensemble perturbed-physics experiments made possible with distributed computing and new adjoint-based estimates of the recent ocean state together with uncertainty therein, we aim to identify free-running versions of an atmosphere-ocean general circulation model that, for the first time, actually reproduce the full evolution of the large-scale climate system over the past 15 years. The ensemble of such models will provide unique insights into the origins, nature and predictability of recent changes in ocean state, together with a valuable tool for assessing future predictability and the risk of substantial Atlantic Meridional Overturning Circulation changes in the longer term. Coupled models currently used both for decadal prediction and longer-term projections of the response to changing boundary conditions typically rely on the comparison of model anomalies from the model climatology with observed anomalies from some estimate of the 'real world' climatology. This is a fundamental problem when either (a) the response to external forcing is uncertain and comparable to any predictability that may arise from the initial state or (b) the system contains significant non-linearities that are likely to impact on any forecast. We will use an entirely novel approach to initialising coupled models directly from a state-of-the-art ocean analysis, using direct perturbation of coupled model parameters to find model-versions that track the real world over the past 15 years. This very challenging objective is made feasible by the unprecedented computing resources, allowing multi-thousand-member ensembles with a fully coupled atmosphere-ocean general circulation model, provided by distributed computing. Our approach will be to initialize tens to hundreds of thousands of perturbed versions of two AOGCMs from the ECCOc ocean analysis, perturbed to allow for both observational and structural uncertainty, estimated from the discrepancies between ECCOc and other analyses. We will use the statistical techniques of likelihood profiling and importance sampling to identify parameter/analysis combinations that allow models to continue to 'shadow' the analysis initially for six months and subsequently, as we home in on promising perturbations, out to the full 15 years. The models used will be HadCM3, which is already set up for distributed computing applications, and a new model based on coupling the HadAM3P model to the MITgcm used in the ECCOc analysis, exploiting information on the parameter-sensitivities of both models that is already available from past ensemble experiments and (in the case of the MITgcm) from the model adjoint. Successful model-versions will then be run free over the full 20th century (for HadCM3) or from 1975 onward (for HadAM3P/MITgcm) to assess the range of AMOC trends they generate in response to total external forcing and anthropogenic forcing only over the past two decades. They will also provide an range of initial conditions for an ensemble prediction experiment to be performed by the VALOR consortium. In addition to its scientific benefits, this project will provide a significant public outreach opportunity, allowing participants to see RAPID data being used directly to address problems of clear and immediate concern.