| 21 Sep 2026
Gaussian-process fusion of ICESat-2 and SMOS/SMAP data for daily estimates of Arctic sea ice thickness and volume
Abstract. We present IS2SMGPSIT-V1, a full-coverage, pan-Arctic, 25 km ×25 km sea ice thickness and volume dataset with daily output spanning seven winter growth seasons between 2018 and 2025. The product fuses ICESat-2 laser-altimetry data with SMOS/SMAP L-band radiometry using Gaussian-process (GP) regression implemented with the open-source GPSat package. ICESat-2 provides high-spatial-resolution thickness estimates along defined ground tracks, whereas SMOS and SMAP provide extensive daily coverage of thin ice (≤0.5 m) thickness. Using each observation’s acquisition time and location reduces sampling artifacts associated with existing monthly binning approaches and fills unobserved grid cells, particularly during early-fall freeze-up. Inner-Arctic-Ocean mean thickness seasonal cycles are broadly consistent with the established IS2SITMOGR4 (ICESat-2 Winter Arctic Monthly Gridded Sea Ice Thickness) Version 4 product, while the fused product extends coverage where the monthly product is sparse. Comparisons with upward-looking sonar measurements from the Beaufort Gyre Exploration Project moorings over 2018–2024 show improved monthly skill relative to IS2SITMOGR4-V4 (r2 = 0.91 versus 0.83; RMSE = 0.19 versus 0.24 m), primarily through improved representation of thin ice. De-seasonalised comparisons show limited sub-monthly skill, indicating that further development and assessment are needed to improve representation of short-timescale variability. The full spatial coverage supports consistent pan-Arctic volume estimates and analysis of interannual anomalies. Version v0.1 of the monthly NetCDF archive is preserved on Zenodo (https://doi.org/10.5281/zenodo.22147361; Petty, 2026b). The data are also hosted on Amazon Web Services S3 as monthly NetCDF files, a single Zarr store, and a version-controlled Icechunk repository to enhance accessibility.
This manuscript presents a daily, 25 km pan-Arctic sea ice thickness and volume dataset. This dataset was created by combining ICESat-2 ice thickness with SMOS/SMAP L-band thin ice thickness within a Gaussian process regression model. The dataset spans seven winter growth seasons from November 2018 to April 2025. As a data paper, it is a valuable contribution. The introduction and discussion are readable for non-specialists, but the other sections use so many acronyms, making it difficult to understand for non-specialists.
1 The manuscript notes that GPSat was "first demonstrated through the fusion of multi-mission radar altimetry data" and that CS3SMOS continues the OI line of Ricker et al. (2017), but it does not clearly distinguish between what is new in the method (i.e. the transfer of GPSat to a new sensor pair versus new GP modelling) and what is new in the product.
2 The title and abstract state that IS2SMGPSIT-V1 provides "daily estimates" and a "daily output" product. However, the validation shows limited sub-monthly skill. The product's anomaly amplitude is only 64–76% of the observed ULS variability, and the fitted temporal length scales are around 35–45 days. This suggests an effective temporal resolution closer to monthly than daily. While the manuscript is transparent about this in the body, recommending temporal aggregation, the headline framing still advertises daily capability.
3 The manuscript reports an improvement in monthly performance compared to IS2SITMOGR4-V4. However, the comparator shares most of its inputs with the fused product. As IS2SITMOGR4-V4 and IS2SMGPSIT-V1 use the same Release 007 ATL10 total freeboards, NESOSIM v1.1 snow loading and ice and seawater density assumptions (Sects. 2.1 and 2.4), the comparison is only partly independent.
4 The manuscript situates IS2SMGPSIT-V1 against CS3SMOS, "the closest existing analogue," and reports agreement. However, two dates are insufficient to determine the positioning of the new product relative to the established CryoSat-2/Sentinel-3/SMOS line, as the authors acknowledge. As it stands, the comparison only establishes that the products exhibit similar spatial patterns on those dates. However, the difference maps show large positive and negative differences. (±1 m). The manuscript also does not test the original claim that laser-plus-L-band GP fusion meaningfully complements or outperforms the radar-altimetry OI line.
5 The manuscript contains many acronyms, such as IS2SITMOGR4, IS2SMGPSIT-V1, IS2SITDAT4, IS2SITMOGR4-V4, NESOSIM, BGEP and ULS. The product short name is also difficult for non-specialists to understand and remember. I recommend providing a short glossary or using more readable acronyms.