Analyzing two decades of basal melting rates below Pine Island Ice Shelf using multi-sensors remote sensing data

Ice shelves play a crucial role in buttressing the ice flux from the ice sheet. It is therefore critical to monitor their evolution and weakening. Changes in basal melting rates, driven by enhanced advection of circumpolar deep water, are one of the primary drivers of ice shelf weakening in Antarctica. In the Amundsen Sea Embayment, basal melting rates are the highest in Antarctica, exceeding 100 m/yr at the grounding line of Pine Island Glacier (PIG), which has discharged more than 130 Gt/yr of ice into the ocean since 2008 (Mouginot et al., 2013; Rignot et al., 2019). Previous studies of basal melting below PIG (Adusumili et al., 2022) were limited by low spatial resolution, showing significant differences from local high-resolution estimates (Shean et al., 2019), and temporal discrepancies when compared to in-situ ocean observations, highlighting the limitations of existing products (Dutrieux et al., 2014; Jenkins et al., 2018). Biases may arise from the challenge of performing remote sensing in Antarctica (fast changing ice bodies), with sensor-specific complexities (e.g., radar altimetry, laser, stereo-photogrammetry) and reliance on model outputs (SMB, firn). In this study, we revisit the estimation of melting rates on PIG using high-resolution multi-sensor optical imagery from 2000s onward. Leveraging modern geospatial formats like GeoParquet, coupled with DuckDB and high-level tools such as Xarray/Dask, we develop a high-performance pipeline to process heterogeneous elevation datasets. Data from GeoEye/WorldView (Maxar) as well as from ASTER (NASA/METI) were used, regenerated and aligned to a combination of measurements with a centimetric precision from the LVIS and ATM instruments aboard NASA's Operation IceBridge, and the ICESat missions. Dozens of millions of data points are uniformly filtered, advected, and corrected for tides, atmospheric pressure, geoid, and mean dynamic topography throughout the entire observation period with dynamically evolving ice shelf geometry from updated grounding lines and ice front positions. We estimate basal melting on summer mosaics, within a consistent Lagrangian framework by calculating changes in thickness, SMB, firn, and rapid ice advection (Shean et al., 2019; Millan et al., 2023). We quantify the error in melting rates using error propagation and supplement this analysis using different firn and SMB products (RACMO, FDM, CFM). We compare the spatial and temporal variability of our melting rate estimates with previous satellite data on PIG as well as in-situ measurements (Dutrieux et al., 2014). The methodology we propose here is based on state-of-the-art tools in geospatial analysis and offers new perspectives for mapping the evolution of basal melting at high resolution on a regular basis over the past two decades. It provides a coherent framework, with the most precise spatio-temporal measurements, limiting sensor-specific biases, which will be extended to all Antarctic ice shelves. We also provide a more conservative uncertainty estimates based on measurement errors as well as an ensemble-based approach for firn and SMB, which are significant sources of uncertainty. This data will be of direct interest for reanalyzing the stability of ice shelves and for constraining ocean models to better resolve basal melting variability.