A Statistics Program at the National Center for Atmospheric Research

美国国家大气研究中心（NCAR）的地球物理统计项目（GSP）为解决地球科学和统计学之间的重要研究问题提供了一个独特的机会。统计研究包括对非平稳空间和时空数据的分析，计算机实验的设计和分析，过滤，极端统计，数据挖掘和统计计算的贡献。这些方法适用于大型数据集，并且有公开可用的软件来实现空间数据和极值的新方法。对地球科学应用的智力贡献同样广泛，包括引入亚网格尺度过程的随机建模，极端事件的气气学估计，空间相干天气发生器的构建以及数据同化的统计改进方法。本世纪的一项重大科学挑战是了解大气、海洋、陆地过程、生物圈和人类活动之间复杂的相互关系，这些关系决定了地球系统。与此同时，我们还需要预测环境的变化，并将这些变化转化为更直接的经济和社会影响。统计科学也面临着相应的挑战，即如何处理庞大而复杂的数据集，这些数据集现在已成为许多科学和工程领域的常态，并利用传统数值模型提供的丰富结构和先验知识。通过对年轻研究人员的培训，积极的访问计划和参与NCAR的几个战略计划，这项工作达到了地球科学统计学家和研究人员的广泛社区。在这种情况下，一些具体的问题是设计和分析模拟地球气候的大型计算机实验，将历史气象数据集外推到没有观测到的地方，改进用于进行数值天气预报的算法，以及识别包含随机过程以表示复杂地球物理现象的模型。贯穿本研究的一个共同因素是对衍生方法或结果中的不确定性进行量化。不确定性估计通常很难在没有统计处理的情况下获得，但当分析随后用于决策或评估风险时，不确定性估计尤为重要。

ABSTRACTPI: Douglas Nychkaproposal: 0355474The Geophysical Statistics Project (GSP) at the National Center forAtmospheric Research (NCAR) provides a unique opportunity to addressimportant research problems at the interface between the geosciences andstatistics. The statistical research includes contributions to theanalysis of nonstationary spatial and spatio-temporal data, the design andanalysis of computer experiments, filtering, the statistics of extremes,data mining, and statistical computing. These methods are adapted to largedata sets and publicly available software is available to implement newmethods for spatial data and extremes. The intellectual contributions toapplications in the geosciences is equally broad and includes theintroduction of stochastic modeling for sub-grid scale processes,estimating the climatology for extreme events, the construction ofspatially coherent weather generators, and statistical improvements tomethods of data assimilation.A grand scientific challenge for this century is to understand the complexinterrelationships among the atmosphere, ocean, land processes, biosphereand human activities that define the Earth System. Coupled with thiseffort is the need to predict changes to our environment and to translatesuch changes into more immediate economic and societal impacts. There isa corresponding challenge for statistical science to tackle the large andcomplex data sets that are now the norm in many areas of science andengineering and to leverage the rich structure and prior knowledgeafforded by traditional numerical models. Through the training of youngresearchers, a vigorous visitor program and participation within severalof the NCAR strategic initiatives, this work reaches a wide communityof statisticians and researchers in the geosciences. In this context, somespecific problems are the design and analysis of large computerexperiments that simulate the Earth's climate, extrapolating historicalmeteorological data sets to locations where they are not observed,improving the algorithms used to make numerical weather predictions, andidentifying models that include random processes to represent complicatedgeophysical phenomenon. One common element throughout this research isquantifying the uncertainty in a derived method or result. Uncertaintyestimates are often difficult to obtain without a statistical treatmentbut are particularly important when the analysis is subsequently used fordecision making or to assess risks.