Validation of ITS_LIVE v2.0 mountain glacier velocities using in-situ GNSS data

Abstract. Glacier surface velocity is receiving increasing attention as it provides essential information on glacier dynamic processes, mass balance estimation, and glacier contributions to sea-level rise under global warming. Velocity products derived from satellite remote sensing have been widely applied to investigations of glacier surging dynamics, ice discharge quantification, and regional hydrological assessments. However, systematic in situ validation of satellite-derived velocity products over small mountain glaciers remains limited, with few studies providing comprehensive error assessment and uncertainty quantification. This gap hinders the reliable application of such products in regional glacier dynamic assessments and introduces unquantified uncertainties into downstream analyses, including mass balance estimation and ice discharge calculations. Here, we validate ITS_LIVE version 2 surface velocities derived from Sentinel-1, Sentinel-2, and Landsat-8 against in situ GNSS measurements over nine glaciers spanning three regions with contrasting dynamic regimes: the Yukon, Canada (fast-moving glaciers), the French Alps (narrow valley glaciers), and the Swiss Alps (slow-moving, small-scale glaciers). For the Yukon region, we validate ITS_LIVE image-pair time series against dense daily GNSS observations, applying two outlier removal strategies — error thresholding and spatial filtering — to improve data quality. For the European Alps, we validate both simple annual averages of ITS_LIVE image-pair products and ITS_LIVE annual composite products against annual-interval GNSS measurements. Regarding the validation results in the Yukon region, Sentinel-1 exhibits lower error in the range direction than in the azimuth direction, with the vy component (RMSE 9.4 m/yr, bias −1.7 m/yr) outperforming the vx component (RMSE 23.1 m/yr, bias 0.8 m/yr), reflecting the inherent anisotropy of SAR measurements. Among optical sensors, Landsat-8 exhibits lower spread (RMSE 16.3 m/yr, NMAD 4.3 m/yr) but higher systematic bias (3.7 m/yr), while Sentinel-2 shows greater spread (RMSE 22.4 m/yr, NMAD 9.6 m/yr) but negligible bias (−0.1 m/yr). Both optical sensors successfully detect the 2022 Lowell glacier surge event, with velocities exceeding 500 m/yr, demonstrating the capability of ITS_LIVE products to capture dynamic glacier flow events. Owing to its shorter repeat cycle, Sentinel-2 resolves finer temporal variability and retrieves higher surge velocities (>600 m/yr), though with increased uncertainty at shorter temporal baselines. As for the validation results in the European Alps, ITS_LIVE annual composite products are validated against annual-interval GNSS measurements. The annual composite products show lower bias (−1 m/yr in Switzerland and −29 m/yr in France) compared to simple annual averages of ITS_LIVE image-pair products (bias of 5–15 m/yr in Switzerland and −28 to 40 m/yr in France), demonstrating the effectiveness of the error-weighted least-squares fitting approach used in composite generation. Additionally, we validate the velocity errors provided with the ITS_LIVE products against GNSS–ITS_LIVE deviations and examine how velocity uncertainty evolves with image-pair time separation. Overall, the ITS_LIVE reported errors agree well with GNSS–ITS_LIVE deviations. This study bridges ground-truth observations with satellite-derived data, offering practical guidance on the applicable scope, expected accuracy, and noise removal strategies for ITS_LIVE v2 over small mountain glaciers.