Collaborative Research: Flexible Statistical Models to Blend Massive Geostationary-Derived Climate Data Records

陆地表面反照率是陆地表面反射的入射太阳辐射的比例。它是全球气候观测系统确定的一个基本气候变量。来自美国、欧盟、日本、瑞士和韩国的机构参与了一项国际努力，致力于从五颗不同的卫星上获取地球静止图像。这些卫星有重叠的区域，两种不同的反照率测量都被反演。该项目将开发混合的全球反照率产品，说明不同反演值之间的差异，并使用概率量化结果反照率值中的不确定性。该项目将利用两个PI的专业知识来开发最先进的方法，可以捕获从地球同步卫星星座观测到的气候变量的时间和空间变异性，如云特征、风速和风向、积雪和降雨量等。该项目将贡献地统计学方法和基于模型的方法来分析、内插和推断多变量空间场，其特征是在连续空间中索引的非平稳空间过程。多亏了局部邻域调节结构，实现了可伸缩性。计算将快速和并发地执行，通过分层结构满足全局一致性。该方法的基于模型的性质允许考虑复杂的观测误差，执行非齐次多变量空间回归，并合并在不同空间聚集水平上获得的观测。时空模型的扩展是无缝获得的，在一个简约的分层模型中结合了时空交互作用。这一奖项反映了NSF的法定使命，并通过使用基金会的智力优势和更广泛的影响审查标准进行评估，被认为值得支持。

Land surface albedo is the fraction of incoming solar radiation reflected by the land surface. It is an essential climate variable as identified by the Global Climate Observing System. An international effort involving institutions from the USA, European Union, Japan, Switzerland and Korea is dedicated to obtaining geostationary images from five different satellites. Those satellites have overlapping areas for which two different measures of albedo are retrieved. The project will develop blended global albedo products that account for the discrepancies between the different retrievals and quantify the uncertainty in the resulting albedo values using probabilities. The project will leverage the expertise of two PIs to develop state of the art methods that can capture the variability in time and space of climate variables that are observed from a constellation of geostationary satellites, like cloud characteristics, wind speed and direction, snow cover, and precipitation, among others.The project will contribute geostatistical methods with a model-based approach to analyze, interpolate and make inferences for a multivariate spatial field, featuring a non-stationary spatial process, indexed in a continuous space. The scalability is achieved thanks to a local neighborhood conditioning structure. Computations will be performed fast and concurrently, satisfying global coherence through a hierarchical structure. The model-based nature of the proposed approach allows to account for complex observational errors, perform non-homogeneous multivariate spatial regression, and combine observations obtained at different levels of spatial aggregation. Extensions to spatio-temporal models are seamlessly obtained incorporating space-time interactions within a parsimonious hierarchical model.This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.