Item

Abstract

Measuring ice sheet elevation changes is essential to monitor ongoing ice-sheet mass loss and related sea-level change and calibrate mass-loss prediction models. Moreover, they are also critical for understanding ice-sheet processes. Different sensors, including laser and radar altimetry and stereo imagery, can capture the change in ice sheet topography and thickness. However, the different resolutions and uncertainty of these sensors and data acquisition missions make the reconstruction of surface change challenging. Laser altimetry provides accurate elevation measurements; however, the data are sparsely distributed in space with temporal gaps due to limited mission extent and atmospheric conditions. These gaps can be filled with radar altimetry data and spatially densified by DEMs. Here we present a fusion framework for combining laser and radar altimetry data and DEMs collected over 50 years. The fusion of the different observations starts with building a correlation matrix, which is then used in a spatiotemporal multivariate sequential Gaussian simulation (SM-SGS) for modeling fused data posterior uncertainty distribution under a Bayesian approach. The latter process is implemented in training regions with good DEM coverage. Because of their dense spatial coverage, the DEMs overcome the sparse spatial distribution of laser data. The outcomes will feed a deep neural network whose training algorithm looks for a model that preserves the measurements and the posterior spatiotemporal uncertainty distribution from the SM-SGS. Using this method, we will combine all observations into an improved spatiotemporal reconstruction to reveal new patterns in the ice-sheet mass-change rate and refine their contribution to sea level rise.