Remote sensing for Antarctic geological mapping: Evidence, validation challenges and future directions

Antarctica remains one of the most difficult regions for geological investigation as extensive ice cover, severe environmental conditions, and logistical constraints continue to limit direct field mapping. This structured critical review brings together published remote-sensing studies that have addressed geological, lithological, mineralogical, and exposed bedrock mapping across Antarctica. Using a transparent and comparative review framework, the paper evaluates the sensors, processing methods, validation approaches, and methodological limitations reported in the available literature, without attempting a quantitative meta-analysis. The reviewed evidence shows that multispectral imagery, especially Landsat and ASTER, forms the strongest and most consistent body of Antarctic geological remote-sensing research. Landsat has mainly supported continent-scale exposed rock mapping and broad lithological interpretation, while ASTER has been used more often for alteration mineral and phyllosilicate mapping because its visible near infrared (VNIR), shortwave infrared (SWIR), and thermal infrared (TIR) bands are better suited to detecting several diagnostic mineral features. Multisensor approaches and hyperspectral data provide valuable spectral, structural, and geochemical information, but their direct use in Antarctic geological studies remains limited and geographically uneven. Across the reviewed studies, the main recurring challenges include spectral mixing, topographic shadow, snow and ice contamination, low solar illumination, limited field validation, spatially biased reference data, and the difficulty of distinguishing spectrally similar lithologies. The review further shows that reported accuracy metrics are not directly comparable across studies unless they are interpreted in relation to sensor type, mapping scale, class definition, validation independence, and geological context. Machine learning, deep learning, explainable artificial intelligence (XAI), and emerging hyperspectral and radar missions offer promising directions for future Antarctic geological remote sensing, although their direct application remains constrained by small labeled training datasets, class imbalance, seasonal transferability, and limited independent validation. To support more reproducible and uncertainty-aware interpretation, this review proposes an integrated workflow linking data selection, preprocessing, feature extraction, geological mapping, uncertainty assessment, validation, and final map production. The review therefore highlights the need for stronger validation, clearer uncertainty reporting, reproducible preprocessing, and careful separation between methods already tested in Antarctic environments and those that remain prospective directions for future research.