km) of gravity recovery and climate experiment (GRACE) and a 11-month data gap with GRACE follow-on (GRACE-FO) limits applications at the individual ice sheet drainage basin scale and complicates the evaluation of regional ice sheet mass changes. While numerous works have downscaled GRACE-estimated water storage, research on downscaling ice mass change in Antarctica is limited. This study employs joint partial least-squares regression (PSLR) and support vector machine (SVM) method to reconstruct GRACE-derived spatiotemporal data for the Antarctic ice sheet (AIS). The pixel-temporal downscaling (PTD) of random forest (RF) and pixel-spatial downscaling (PSD) of multiscale geographically weighted regression (MGWR) enhance spatial resolution of ice mass changes from 0.25° (~120 km) to 1.92 km. The downscaled results show consistent temporal variation and reduced noise compared to other reconstruction methods. Both RF and MGWR results exhibit high consistency with original GRACE data, with MGWR achieving a correlation coefficient (CC) of 0.99. The MGWR model effectively captures finer signals related to ice flow velocity. When compared to independent free air gravity anomalies, MGWR outperforms RF with improvements of 41.51% and 56.25% in mean correlation for group 1 and group 2 observation points, respectively. In addition, MGWR shows improvements of 16.90%/29.69% for flight Line A and 11.84%/19.72% for flight Line B compared to RF and original GRACE results. The enhanced spatial resolution offers valuable insights into ice dynamic changes within the Western AIS and Eastern AIS and smaller regions such as the Antarctic Peninsula.">
Improving the Spatial Resolution of GRACE-Derived Ice Sheet Mass Change in Antarctica (original) (raw)