Bayesian Latent Variable Co-kriging Model for Quality Flagged Remote Sensing Observations

Remote sensing data products often include quality ags that inform users whetherthe associated observations are of good, acceptable or unreliable qualities. However,such information on data  delity is not consistently considered in remote sensing dataanalyses. Motivated by observations from the Atmospheric Infrared Sounder (AIRS)instrument on board NASA's Aqua satellite, we propose a latent variable co-krigingmodel with separable Gaussian processes to analyze large quality-agged remote sens-ing data sets together with their associated quality information. We augment theposterior distribution by an imputation mechanism to decompose large covariance ma-trices into separate computationally e cient components taking advantage of theirinput structure. Within the augmented posterior, we develop a Markov chain MonteCarlo (MCMC) procedure that mostly consists of direct simulations from conditionaldistributions. In addition, we propose a computationally e cient recursive predictionprocedure. We apply the proposed method to air temperature data from the AIRS in-strument. We show that incorporating quality ag information in our proposed modelsubstantially improves the prediction performance compared to models that do notaccount for quality ags.

遥感数据产品通常包含质量标记（quality flag），用于告知用户相关观测数据的质量等级为优良、可接受或不可靠。然而，在遥感数据分析中，此类关于数据保真度（data fidelity）的信息并未得到一致的重视与应用。依托搭载于美国国家航空航天局（NASA）Aqua卫星上的大气红外探测器（Atmospheric Infrared Sounder, AIRS）的观测数据，我们提出一种结合可分离高斯过程的潜变量协同克里金模型，用于对带有质量标记的大规模遥感数据集及其配套质量信息开展联合分析。我们通过插补机制对后验分布进行扩充，利用输入数据的结构特征，将大规模协方差矩阵拆解为多个独立的、计算高效的子组件。在扩充后的后验分布框架下，我们构建了一套马尔可夫链蒙特卡洛（MCMC）推断流程，其核心步骤大多为基于条件分布的直接抽样。此外，我们还提出了一种计算高效的递归预测方法。我们将所提方法应用于AIRS传感器获取的气温数据中。实验结果表明，相较于未考虑质量标记的模型，在我们提出的模型中融入质量标记信息后，预测性能得到了显著提升。