Collaborative Research: Scalable Statistical Validation and Uncertainty Quantification for Large Spatio-Temporal Datasets

计算机模拟、卫星和各种其他技术进步通过产生大量以前无法收集的数据集，为进入前所未有的科学领域铺平了道路。正确地利用这些新的数据产品来促进科学发现，需要它们首先(i)得到验证，（ii）与适当的不确定性测量相匹配。验证数据产品需要将“合成”数据（例如，计算机模型模拟、遥感测量或统计预测）与观测对应数据进行比较，以证实数字数据用于科学发现。不确定性量化（UQ）是验证的必要组成部分，需要考虑和说明与来自数字或观测数据的科学结论相关的不确定性。本研究的目的是通过发展统计方法来进行验证和不确定度量化，促进使用数字数据产品的科学发现。考虑到执行统计验证和UQ的强烈需求和面临的重大挑战，本研究将(i)开发基于科学动机特征的模拟和数字数据集的新验证策略；（ii）开发多元时空统计模型，可用于实施和执行时空数据产品的UQ；并且（iii）开发可扩展的计算技术来拟合已开发的时空统计模型。这项研究将在大气、农业和环境科学的数据产品上实施这些技术，以促进它们在科学调查中的使用。

Computer simulations, satellites and various other technological advances have paved the way into unprecedented scientific territory by generating volumes of previously uncollectable datasets. Properly utilizing these new data products to promote scientific discovery requires that they, first, be (i) validated and (ii) paired with an appropriate measure of uncertainty. Validating a data product entails comparing "synthetic" data (e.g., computer model simulations, remote sensing measurements, or statistical predictions) with observational counterparts to substantiate the digital data for its use in scientific discovery. Uncertainty quantification (UQ) is a necessary component to validation and entails accounting for and stating the uncertainties associated with scientific conclusions derived from digital or observational data. The purpose of this research is to promote scientific discovery using digital data products by developing statistical methods to perform validation and uncertainty quantification.Given the strong need to perform, and the substantial challenges facing, statistical validation and UQ, this research will (i) develop new validation strategies for simulated and digital datasets based on scientifically motivated features; (ii) develop multivariate spatio-temporal statistical models that can be used to implement, and perform UQ for spatio-temporal data products; and, (iii) develop scalable computation techniques for fitting the developed spatio-temporal statistical models. This research will implement these techniques on data products in atmospheric, agricultural and environmental sciences to facilitate their use in scientific inquiry.