BAYESIAN HIERARCHICAL MODELS FOR HIGH-DIMENSIONAL REMOTE SENSING DATA WITH APPLICATION TO GRAVITY RECOVERY AND CLIMATE EXPERIMENT (GRACE) TOTAL WATER STORAGE ANOMOLIES

Satellite remote sensing is a valuable source of observations of hydrologic variables, especially over large regional scales and in areas with sparse in-situ data. Gravity anomaly measurements from the NASA/German Space Agency’s Gravity Recovery and Climate Experiment (GRACE) have been used to estimate temporal variations in groundwater, surface water, and soil moisture storage at regional scales world-wide. GRACE observations are available as an unevenly spaced time series at coarse resolution (approximately monthly) due to orbit limitations. Currently, data are not collected when the satellites are eclipsed and cannot use solar energy due to degraded battery capacity, leading to gaps in the dataset. These factors complicate the use of GRACE observations in hydrologic models and scientific studies, where regularly spaced data is desirable. Our study evaluates methods to model the GRACE observations within a Bayesian framework. We use a spatio-temporal mixed effects model to estimate missing values, derive a regularly spaced time series, and forecast. Spatio-temporal data such as the GRACE observations are extremely high dimensional, which leads to practical issues when formulating statistical models. Our study evaluates Moran’s I and empirical orthogonal basis functions to reduce the high-dimensional parameter space. Overall, it is hoped that these methods will facilitate the future use of GRACE observations in hydrological models and scientific studies.