Advanced computing architecture to support the estimation and reporting of UK GHG emissions

温室气体（GHG）的排放可以通过使用计算要求很高的贝叶斯“反”方法从大气浓度的测量中推断出来。布里斯托尔大学（UoB）和爱丁堡大学（UoE）的研究小组正在利用这些信息a)量化英国和其他国家排放的幅度和不确定性，b)确定大气温室气体自然变率的驱动因素。这项工作是几个主要项目的基础，包括：a)能源和气候变化部（DECC）监测网络，负责向联合国气候变化框架公约报告英国的温室气体排放，b) 300万英镑的nerc资助的英国温室气体和全球排放（GAUGE）联盟（帕尔默是PI，里格比是co-I）， c)美国宇航局和DECC资助的先进全球大气气体实验（里格比是成员），d)国家地球观测中心。这项工作包括两个阶段。首先，化学运输模型（CTMs，例如英国气象局的NAME模型）在多节点集群上运行，然后将其输出与使用（通常）单节点数据分析系统进行排放验证的观测结果进行比较。后者的统计技术涉及在超大数组上使用CPU和内存密集型线性代数算法，这已经突破了我们现有基础设施的极限。DECC网络和GAUGE内部的活动现在提出了进一步的挑战：1)充分利用快速增长的异构测量数据（数百万个数据点）；2)利用这些数据以比以往更高的分辨率推断排放（例如，利用英国上空1.5公里水平分辨率的NAME模式输出）。拟议的资产将有助于加强我们开展这项工作第二阶段的能力。

Greenhouse gas (GHG) emissions can be inferred from measurements of their atmospheric concentration using computationally demanding Bayesian "inverse" methods. This information is being used by research groups at the University of Bristol (UoB) and the University of Edinburgh (UoE) to a) quantify the magnitude and uncertainty of emissions from the UK and other countries, and b) determine the drivers of natural atmospheric GHG variability. This work is underpins several major projects including: a) the Department for Energy and Climate Change (DECC) monitoring network, responsible for reporting UK GHG emissions to the United Nations Framework Convention on Climate Change, b) the £3m NERC-funded Greenhouse gAs Uk and Global Emissions (GAUGE) consortium (Palmer is PI, Rigby is co-I), c) the NASA and DECC-funded Advanced Global Atmospheric Gases Experiment (Rigby is member), and d) the National Centre for Earth Observation. This work involves two stages. Firstly, chemical transport models (CTMs; e.g. the UK Met Office NAME model) are run on multi-node clusters, before their output is compared to observations for emissions verification using (usually) single-node data analysis systems. The statistical techniques for the latter involve the use of CPU- and memory-intensive linear algebra algorithms on extremely large arrays, which are already pushing the limits of our existing infrastructure. Activities within the DECC network and GAUGE now pose further challenges: 1) to fully exploit a rapidly growing quantity of heterogeneous measurement data (many millions of data points); 2) to use these data to infer emissions at higher resolution than ever before (e.g. making use of NAME model output at a horizontal resolution of 1.5 km over the UK). The proposed assets will help to strengthen our ability to carry out this second stage of this work.