| 05 Jun 2026
From snow wetting to pond formation: Stage-resolved L-band sea ice roughness during the Arctic melt season
Abstract. Arctic summer surface evolution is commonly characterized using optical melt pond fraction, but optical retrievals are limited by persistent cloud cover and cannot resolve internal snow–ice changes that occur before ponds become visible. L-band passive microwave observations are largely unaffected by clouds and respond to near-surface dielectric changes within the snow and upper ice layer, suggesting their potential to detect pre-pond surface transitions. This study examines whether small-scale L-band sea ice roughness (L-SIR), retrieved from Soil Moisture Active Passive brightness temperatures, provides a physically consistent precursor to optically detectable melt pond development. Using observations from 2017 to 2023 over a multiyear ice region in the central Arctic, we analyze the temporal ordering among the transition to positive net surface energy flux, the L-SIR transition, and melt pond formation, together with stage-dependent relationships between L-SIR and surface energy balance variables. The L-SIR transition occurs approximately two weeks after the net surface energy flux becomes positive and approximately four weeks before melt pond formation, with this ordering reproduced across years and latitudes. Correlations with surface energy variables also change systematically across three stages, shifting from a temperature-related regime to a latent heat- and longwave-associated regime during active snow ablation and meltwater redistribution, and finally to a pond coverage-related regime. These results indicate that L-SIR provides a complementary satellite-derived product for characterizing pre-pond Arctic summer surface evolution that is not captured by optical melt pond fraction alone.
Overview of the paper
The topic of this research, and what authors tried to accomplish is of great interest. It aims at addressing one big limitation on melt season remote sensing-based studies, that is the fact that melt pond optical methods cannot be used in the precise moment where the surface undergoes significant changes that determine pond onset. This is why approaches that investigate other sides of the EM spectrum, in this case L-band, are of great importance. L-band capabilities offer a promising complementary tool to look at the pre-pond interval. The fact that it is an effort to provide a physically anchored, satellite-observable marker for a transition that previously has little to no continuous observational proxy is significant. Moreover, a great strength is the fact that it covers 7 years, meaning it goes through 7 independent melt seasons. So, while covering only a very narrow region, and a single ice regime (MYI), it is nevertheless a very interesting proof-of-concept, keeping in mind the very specific type of ice it's looking at. However, I would recommend a few changes (in quantity), some of which could be considered major, because being a proof-of-concept and novel approach, it is important that the methodology behind it is strong to ensure that results, and deriving claims, have strong foundations and justifications.
Major revisions suggested:
#1. Ts enters directly into the L-SIR retrieval approach (Eq. 1, as the denominator of the rough-surface reflectivity). The paper then reports the r determined between L-SIR and Ts anomalies in Stage 1, interpreting it as physical. However, during the Discussion, it is mentioned that the correlaction or part may come from the retrieval formulation. Given the importance of this, authors are recommended to undergo a sensitivity test showing how much of the L-SIR–Ts correlation is mechanically induced by the retrieval equation, versus how much reflects independent polarization-contrast information. In other words, to decouple the physical signal from the retrieval’s temperature dependence. Otherwise the Stage-1 result should be framing in a different manner, since right now it is identified as the early precursor mechanism in this work.
#2. The study's central claims rest entirely on a single, fixed analysis domain (84–85.5°N, 50–70°W, near the Lincoln Sea), however, given that the central Arctic spans a wide range of multiyear-ice conditions, and the risk of Interpretating the Lincoln Sea sector as representative of that broader region, would recommend to frame its selection as a best-case (given ice cover e.g.). Moreover, also before claiming generality at the scale, as one could interpreter from the abstract and conclusion, it would be best supported with at least a validation site, or, a well supported justification for why this particular sector (despite the narrow extent) can be expected to generalize. If both cases are absent, the language and statement of this particular sector as a recurring feature of the central Arctic should be softened through the abstract, results, and conclusions so it reflect that the finding is a single-site, single-ice-regime result, however robust across the seven years it was tested.
#3. There is an established literature on detecting melt onset from passive microwave brightness temperatures (typically higher-frequency channels, e.g., 19/37 GHz methods). Given the paper's central claim of providing "complementary" pre-pond information, a comparison (or at least explicit discussion of what is genuinely novel about an L-band roughness-based precursor vs. conventional microwave melt-onset detection) is needed so it allows the paper to position the L-SIR onset relative to these widely used melt-onset products.
#4. Some components on the methodology would need a moe clear description, also to allow results interpretation: a) Could the authors explicitly describe how are the weekly MPD2 composites converted to the daily resolution used in the time-series analysis and the threshold-crossing date definition?; b) Provide more detail on the criteria of exclusion of ‘grid cells classified into the same stage on a given date’; c) Specify the resampling method from SMAP grid to 12.5km and finally, d) The correlation values (r) reported in Figures 5 and 6 are calculated from daily data that has been smoothed and is therefore not fully independent from one day to the next (autocorrelation). This matters because when you treat each day as if it were a separate, independent data point, it makes correlations look more statistically reliable than they actually are. With 21 different correlation values reported across the three stages and seven variables, the authors should show some way of checking which of these correlations are actually statistically solid.
Minor revisions suggested (not by order):
#1. The "across years and latitudes" claim is based on a single, narrow domain (1.5° latitude × 20° longitude) over multiyear ice. This is explicitly acknowledged as a limitation, but a reviewer is likely to ask for at
#2. Figure 3 (and similarly Figure 6's correlation matrix) should be revised using a colorblind-safe categorical palette, consistent with Copernicus' own accessibility guidance for color figures. The current scheme pairs red and green as distinct, non-adjacent categories, but I guess that the editors from the journal could best advise on this.
#3. Section 2.3 states that "Quality assessment flags were applied to retain only observations of acceptable quality," but does not specify which SMAP QA flags were used. This should be stated explicitly for reproducibility.
#4. No retrieval uncertainty is reported for L-SIR itself. Given that the stage-wise ΔL-SIR values (~0.18–0.39 cm) are modest relative to the retrieval's full dynamic range (~0–2 cm), even an approximate propagated uncertainty (from TB and Ts input uncertainties) would help readers judge whether the reported stage transitions are clearly above the retrieval noise floor.
#5. Page 14: an incomplete/awkward phrase in "Whether the L-SIR onset can be used predictively to estimate MPF onset will need to be tested in future work using broader regions, sea ice type." This reads as though a word was dropped — likely intended as "...using broader regions and different sea ice types."
#6. Caution with references: in the text there is "Lange et al., 2018" but in reference list is dated 2019. Recommend to update "Aparicio et al., 2023" which is a preprint to the most recent version (2026).
#7. Duplicated word, Sect. 2.3"daily vertically and horizontally polarized polarized brightness temperatures"
#8. Add significance markers (e.g., asterisks or a p-value table) to the Fig. 6 correlation matrix.
#9. The Conclusions section largely restates the Discussion; perhaps could be summarized with key points to avoid redundancy/repitition.