Calving front positions for Greenland outlet glaciers (2002–2021): a spatially extensive seasonal record and benchmark dataset for algorithm validation

Abstract. Calving front positions of marine-terminating glaciers are a key indicator of variations in glacier dynamics, ice–ocean interactions, and serve as critical boundary conditions for ice sheet models. High-precision, long-term records of calving front variability are essential for understanding glacier recession and calving processes, improving mass loss estimates, and supporting the development and validation of robust automated front-tracking algorithms. However, existing datasets often exhibit limited spatial coverage, inconsistent temporal resolution, and heterogeneous delineation methods, which result in variable accuracy and insufficient detail, reducing the performance and transferability of automated calving front detection. Here, we present a spatially extensive, high-accuracy dataset of glacier calving front positions across Greenland, intended as a benchmark for algorithm training, model–data integration, and studies of seasonal glacier dynamics. The dataset comprises approximately 12,000 manually delineated calving front positions for ~290 outlet glaciers from 2002 through 2021, extracted from multi-source satellite imagery (Landsat, Sentinel-1/2, MODIS, ENVISAT, and ERS). Delineations were conducted using standardized workflows in the Google Earth Engine platform and ArcGIS, and each record is accompanied by comprehensive metadata, including acquisition date, digitization method, source imagery, and other relevant attributes. Positional accuracy was evaluated through comparison with high-resolution PlanetScope imagery and manually interpreted reference datasets, confirming high geometric fidelity with positional offsets ranging from about 40 to 100 m across representative glaciers, depending on image resolution and terminus complexity. In contrast, automated products tend to show reduced accuracy in verification areas with complex terminus morphology, reflecting their high sensitivity to image quality, limited generalizability across heterogeneous geometries, and the absence of large-scale, high-precision training data. This dataset contributes to mitigating these challenges by providing dense, manually validated, high-precision observations across Greenland, serving as a robust benchmark for developing and validating automated front detection algorithms, refining boundary representations in ice sheet models, and advancing understanding of ice–ocean interactions. The dataset is publicly available at https://doi.org/10.5281/zenodo.16879054 (Xi et al., 2025).