| 09 Feb 2026
Reconstructing Glacier Dynamics in Complex Terrain with ICESat-2 and Gaussian Process Interpolation
Abstract. We apply Gaussian Process Regression (GPR) to ICESat-2 along-track height change data to generate spatially continuous glacier height change fields across two glaciated regions with complex topography and dynamic behaviour: Larsen-B (Antarctic Peninsula) and central Southern Svalbard. For Larsen-B, GPR-derived height change rates from 2021–2024 average −0.61 ± 0.02 m a–1, corresponding to a volume loss of 4.55 ± 0.17 km³ a–1, similar to independent estimates from TanDEM-X. In Svalbard, we observe widespread thinning (−1.57 ± 0.03 m a–1) and detect clear signals of surging glaciers. GPR's multi-dimensional, uncertainty-aware framework enables accurate interpolation across data gaps and supports the detection of localized dynamic events, such as surges. Sensitivity tests show that interpolation errors increase with gap size, slope, and extrapolation distance. Our findings demonstrate that GPR is well-suited for enhancing the spatial and temporal resolution of altimetry-based glacier monitoring, enabling improved estimates of height and volume change while reconstructing spatially localized phenomena such as surge activity and other transient glacier dynamics.
General Comment:
This manuscript by Seehaus and co-authors presents an application of Gaussian Process Regression (GPR) to interpolate ICESat-2 derived glacier surface elevation change (dh/dt) across two glacierized regions characterized by complex terrain and heterogeneous glacier dynamics: the Larsen-B embayment on the Antarctic Peninsula and central southern Svalbard.
The topic is timely and relevant for the cryosphere community. Satellite altimetry provides highly accurate but spatially sparse measurements, and developing robust approaches to reconstruct spatially continuous elevation change fields remains an important challenge. The use of a probabilistic interpolation framework, such as GPR, is well motivated, and the manuscript demonstrates that the approach can reproduce spatial patterns of glacier thinning and thickening, including signals associated with glacier surge activity.
Overall, the manuscript is clearly written and well structured. The use of two contrasting study regions strengthens the evaluation of the approach, and the sensitivity experiments provide useful insight into the effects of data gaps and terrain complexity. The figures effectively illustrate the spatial patterns of elevation change and interpolation uncertainty. However, several methodological aspects would benefit from clarification or further discussion.
Major Comments:
1. Justification of GRP Kernel
The manuscript adopts a Matérn 5/2 kernel with correlation lengths derived from semivariogram analysis. While this choice appears reasonable, the justification for the specific kernel and parameter values remains somewhat qualitative. It would strengthen the study's methodological rigor if the authors clarified whether kernel hyperparameters were optimized using marginal likelihood within the GP framework or fixed based on the semivariogram analysis. In particular, expanding the discussion of alternative kernel testing and kernel/correlation-length combinations would strengthen the justification for choosing this kernel.
2. Use of Glacier ID as a Predictor
Including the glacier ID as a predictor variable is an interesting approach to reduce spatial leakage between adjacent glaciers. While the examples in the manuscript demonstrate that this approach reduces leakage in practice, the inclusion of this variable also raises some questions. Glacier ID is a categorical variable, rather than a continuous physical variable, like the other predictors. Treating the ID as a numerical feature in the covariance kernel may introduce artificial relationships between glaciers. While this may not be a major concern, it would be helpful for the authors to discuss the statistical implications of this choice and whether alternative approaches were considered.
3. Spatial Correlation Length for Smaller Glaciers
The spatial correlation length used in the model (~11 km) appears relatively large compared to the characteristic scale of many glaciers in the study regions, particularly in Svalbard, where surge-related signals and terminus dynamics may occur over much shorter distances. A brief discussion of how this correlation length affects the model's ability to capture sharp spatial gradients in dh/dt would be useful. In particular, the authors may wish to comment on whether the chosen correlation length could lead to smoothing of localized elevation-change signals.
4. Artificial Gap Experiments
The validation strategy combines comparisons with TanDEM-X elevation change fields and artificial data-gap experiments. These tests are helpful and provide useful insight into the interpolation performance. The artificial gaps used in the sensitivity analysis are circular. whereas real ICESat-2 sampling gaps are typically elongated and aligned with satellite ground tracks. The authors may wish to briefly discuss whether this difference could influence the interpretation of the gap-filling experiments.
Minor Comments: