The Terra-correlator: A computing facility for massive real-time data assimilation in environmental science

1.地震干涉测量：地震学家传统上通过测量、分析和模拟地震自然产生的信号或人为产生的信号来探测地球。自2003年以来，几位作者基于地震图的相互关联得出了新的方法，这种方法允许在没有脉冲事件的情况下从地球环境的波动中获得类似的信息，以前被视为噪声。目前的全球数据集可用于每两小时产生约20KB大小的地球脉冲响应，或每天高达2TB的输出数据，其中大部分提供了全新的信息。这将在节点1上完成。然后将连续并行地对输出进行数据挖掘，以提取关于地下变化的近乎实时的信息，用于预测目的。这需要在单独的节点(节点2)上应用延时模式、模式匹配和其他分析和建模方法中的额外的接近实时的相关性，该节点绑定到一个门户，该门户将确保这种分析和对未来行为或事件的任何预测(例如，地震速度变化后的火山喷发)是在真实事件时间之前以可核查的方式完成的，从而一举消除了基于过去数据分析预测质量时的回溯选择偏差的长期问题。地球系统科学：(A)一项反复出现的挑战是分析直接对地观测、卫星和模型数据，并对不确定性分析和参数空间探索进行数据和计算机密集型处理。这一新设施将用于估计2000-2013年期间的碳储量和通量，并确定其可信区间。最终，我们在英国的1平方公里可能被用来产生对温室气体排放的近乎实时的分析，这可能对政策制定者有用。还将在观测系统模拟实验中探索计划中的地球观测任务的信息内容，如欧空局的生物质飞行任务。(B)由于缺乏记忆力，我们目前对气候模型的性能与结果进行比较的许多典型分析都受到多次读取数据的限制。我们将分析NOAA的数据，以发现比以前更高分辨率的全球25公里极端事件，以研究极端降水事件的机制和特征特征。3.多孔岩石中变形和流体流动的高分辨率实时监测：在一系列应用中，X射线C-T(计算机层析成像)成像的计算量很大，但操作员需要进行大量的后处理来调整结果图像，特别是将孔隙空间与固相分离。时间推移测量为跟踪裂缝或孔隙中的流体流动提供了可能性，或者通过跟踪(互相关联)颗粒运动以前所未有的分辨率观察变形。因此，在前两个应用程序中使用的基础设施开销将用于使这种分析在“现场”实验中成为可能，以支持正在进行的岩石物理实验和建模工作，最初的重点是碳酸盐。

1. Seismic interferometry: Seismologists have traditionally explored the Earth by measuring, analysing and modelling signals generated naturally by earthquakes, or deliberately by man-made sources. Since 2003, several authors derived new methods based on cross-correlation of seismograms that allowed similar information to be obtained in the absence of impulsive events - from the Earth's ambient fluctuations, previously regarded as noise. Current global data sets can be used to produce Earth impulse responses of ~20kB size every two hours, or up to 2TB of output data per day, the bulk of this providing completely new information. This will be done on node 1. The output will then be data-mined continuously and in parallel to extract near-real time information about subsurface changes for forecasting purposes. This requires additional near-real time correlations in time-lapse mode, pattern matching, and other methods of analysis and modelling to be applied on a separate node (node 2), tied to a portal that will ensure such analysis and any forecasts of future behaviour or events (say a volcanic eruption following a seismic velocity change) is verifiably done in advance of the real event time, removing at a stroke perennial problems with retrospective selection bias when analysing forecast quality based on past data.2. Earth system science: (a) A recurring challenge is to analyse direct Earth observation, satellite and model data with data- and compute-intensive processing for uncertainty analyses and parameter-space exploration. The new facility will be used to produce estimates of carbon stocks and fluxes with confidence intervals over the period 2000-2013. Ultimately our UK runs at 1 km2 may be used to generate near-real time analyses of GHG emissions that are likely to be useful for policy makers. The information content of planned EO missions, such as ESA's BIOMASS mission, will also be explored in observing system simulation experiments. ( b) Many of our typical current analyses of the performance of climate models in comparison with the outcome are constrained by reading in data multiple times, due to lack of memory. We will analyse NOAA data to find extreme events at much higher resolution than before of 25km globally, to research the mechanisms and characteristic signatures of extreme precipitation events. 3. High-resolution real time monitoring of deformation and fluid flow in porous rocks: X-ray C-T (computer-tomography) imaging is computationally intensive in a range of applications, but a very large amount of post-processing is required by the operator to tune the resulting image, notably to separate pore space from the solid phase. Time-lapse measurements open up the possibility of tracking fluid flow in fractures or pores, or observing deformation at unprecedented resolution by tracking (cross-correlating) particle movements. Accordingly the infrastructure overhead used in the first two applications will be used to make such analysis possible in a 'live' experiment, in support of on-going experimental and modelling work in rock physics, initially focussed on carbonates.