Adaptation of emerging computational technology for carbon and palaeoclimate modeling

人类文明面临的重大挑战之一是人为造成的气候变化，其中很大一部分是由于燃烧化石燃料向大气中释放二氧化碳造成的。当前的气候科学研究旨在更好地理解在漫长的地球历史过程中大气二氧化碳水平的自然变化与自然气候变化之间的关系。基于深海沉积物岩心的重建表明，在过去的~ 5000万年里，地球气候已经不稳定地从格陵兰岛和南极洲没有冰的温暖“温室”状态过渡到现代的“冰窖”状态，两极附近有大陆规模的冰盖。对这些深海沉积物的其他类型的测量表明，大气中的二氧化碳水平和海洋酸度在同一时间段内普遍下降，从而提高了大气中二氧化碳含量下降可能是5000万年全球变冷的关键原因的可能性。这里提出的工作的中心目标是提高我们对大气中二氧化碳在调节气候系统状态中的作用的理解。由于所涉及的物理、化学和生物过程的复杂性，计算机模型已成为气候研究的核心工具。考虑到我们星球的浩瀚和地球数百万年的历史，很明显，地球系统建模是一项计算挑战，需要使用越来越快的超级计算机。在过去十年和可预见的未来，高性能计算的最大变化是“图形处理单元通用计算”（GPGPU）的发展，它大大减少了给定计算机在给定时间内执行给定计算所需的尺寸和能耗。因此，在容纳超级计算机的建筑物的有限空间和电源供应范围内，可以使用GPGPU解决更大的计算问题，而不是依赖于CPU（中央处理器）硬件的经典方法。然而，现有的地球系统模型是为cpu架构开发的，因此无法充分利用GPGPU的优势。本文提出的工作方法是开发gpu原生模型组件（海洋生物学，物理海洋环流的化学混合，海洋与大气之间的二氧化碳交换等），从而充分利用GPGPU的计算优势。这方面的工作建立在高性能计算和地球科学领域的科学家之间的合作和知识交流的基础上。最终，对地球历史和气候变化的更好理解必须基于沉积记录作为证据的重建。这些观测结果经常与模型模拟进行比较，作为验证某一特定模拟是否与不同类型的数据一致的一种方式。有了这里提出的工作，将有可能系统地计算数千次模拟，然后与观测结果进行细致的比较。基于这一进展，数据和模型之间（以及数据和数据之间）的交叉比较将有助于更好地了解地球系统在过去是如何变化的，以及驱动过程是什么。这些关于过去的知识将为我们对未来气候和碳循环的展望提供信息。

One of the grand challenges to human civilization is man-made climate change, much of which is caused by the burning of fossil fuels that releases CO2 to the atmosphere. Current research in climate science is aiming to better understand the relationship between natural changes of atmospheric CO2 levels and natural climate change over the long course of Earth's history. Reconstructions based on sediment cores recovered from the deep ocean demonstrate that over the last ~50 million years Earth's climate has unsteadily transitioned from warm "greenhouse" conditions with no ice in Greenland and Antarctica to the modern "icehouse" state with continental-scale ice caps near the poles. Other types of measurements on these deep sea sediments indicate that atmospheric CO2 levels and ocean acidity have broadly declined over the same timeframe, thereby raising the possibility that atmospheric CO2 decline may have been a key reason for ~50 million years of global cooling. The central aim of the work proposed here is to improve our understanding of the role of atmospheric CO2 in regulating the state of the climate system.Because of the complexity of the involved physical, chemical and biological processes computer models have become a central tool in climate research. Considering the vastness of our planet and the many million years of Earth history, it becomes clear that Earth System modeling is a computational challenge that requires the use of ever-faster supercomputers. The greatest change in high-performance computing over the last decade and the foreseeable future is the development of "General Purpose computing on Graphics Processing Units" (GPGPU), which greatly reduces the size and energy consumption a given computer needs to do a given calculation in a given amount of time. Thus, within the limited space and power supply of a building that houses a supercomputer much larger computational problems can be solved using GPGPU rather than the classical approach that relies on CPU (Central Processing Unit) hardware. However, the existing Earth System models were developed for CPU-architectures and are therefore unable to fully exploit the benefits from GPGPU. The approach of the work proposed here is to develop GPU-native model components (marine biology, chemical mixing by physical ocean circulation, CO2-exchange between ocean and atmosphere, etc.) that enjoy the full computational benefit from GPGPU. This aspect of the work builds on the collaboration of and the exchange of knowledge between scientists in the fields of high-performance computing and the Earth sciences.Ultimately, a better understanding of Earth history and climate change must be derived from reconstructions based on the sedimentary record as evidence. These observations are frequently compared to model simulations as a way of verifying that a particular simulation is in agreement with data of different types. With the work proposed here it will become possible to systematically compute thousands of simulations, followed by meticulous comparison to observations. Based on this advance, the cross-comparison between data and models (and between data and data) will help to better understand how the Earth System has changed in the past, and what were the driving processes. This knowledge of the past will then inform our outlook for the climate and the carbon cycle of the future.