the Creative Commons Attribution 4.0 License.
the Creative Commons Attribution 4.0 License.
Recent and future changes in rain-on-snow event characteristics across Svalbard
Abstract. Rain-on-snow (ROS) events in Svalbard are becoming a more frequent occurrence during the winter season due to rapid climate warming across the Arctic in recent decades. ROS events have gained increasing attention in recent decades due to their cascading impacts on the physical environment, and terrestrial and marine ecosystems that are impacted by snowmelt. While the frequency of ROS events in Svalbard has been well studied and documented, other characteristics of ROS, such as their duration, intensity and seasonal timing have received less attention. Such characteristics are equally important to quantify due to their potential consequences for the winter snowpack and snow-dependent ecosystems. This study addresses this knowledge gap using the Copernicus Arctic Regional Reanalysis (CARRA) for the present day analysis and km-scale climate projections from a regional climate model for the future period of 2030–2070 under the high emissions scenario SSP5-8.5. For the present climate, the results show significant and increasing trends in all characteristics but confined mainly to low-lying areas of Nordaustlandet and some areas in the east of the archipelago, while no statistically significant trend was found in the southern and western areas which typically exhibit the largest values in all characteristics. Analysis of the future projections showed that the largest changes relative to present day conditions in all ROS characteristics will take place over the mountainous and glaciated areas in the north and northeast of the archipelago, while some low lying western coastal areas will experience a decrease. This reduction is expected to be the result of fewer days with snow, shortening the season where rain can fall on an existing snow cover. Moreover, while ROS has increased most in November and February, the future climate simulation features a substantial increase in ROS events in April, which experiences very few, if any, ROS events in the present climate, which may lead to considerable changes in snow hydrology. Further work could include analysing an ensemble of climate projections for Svalbard to produce a range of ROS scenarios, as well as carrying out a more in-depth analysis of the hydrological impacts associated with the changes in ROS characteristics identified here.
- Preprint
(17381 KB) - Metadata XML
- BibTeX
- EndNote
Status: final response (author comments only)
-
RC1: 'Comment on egusphere-2025-2099', Anonymous Referee #1, 08 Jul 2025
-
AC1: 'Reply on RC1', Hannah Vickers, 14 Sep 2025
Thank you for the review and helpful points raised.
With regards to the concern about the evaluation of the RCM used (HCLIM) there has now recently been published a report by the Norwegian Meteorological Institute which has provides a comprehensive evaluation of HCLIM against CARRA: https://www.met.no/publikasjoner/met-report/_/attachment/inline/b1b9b27c-4086-452f-a8e7-b7bdbeb4053f:996133826c7a1b2d83d35602f425fedbad34f525/MET-report-01-2025.pdf
The report indicates that compared to CARRA, HCLIM tends to produce a slightly cold bias on land over Spitsbergen but a warmer bias over eastern areas due to lack of snow cover on sea ice in ERA5 in the Barents sea region. This could certainly contribute to the differences we observed in the ROS characteristics between CARRA and HCLIM-ERA5, which shows that HCLIM produces lower estimates of ROS over Spitsbergen compared to CARRA but produces much smaller differences over the northeastern areas (eg. Nordaustlandet). Based on this we will amend our discussion to highlight this point in explaining the observed differences.
The 79 simulations in the CORDEX-ARC-22 ensemble includes all variables and all scenarios together. When these are filtered for 2-metre temperature (tas), there are only two RCMs (CanRCM4 and CRCM5) available for the ERA5-driven historical period, and only one RCM (CanRCM4) available for future scenarios. Since these simulations are almost 10 times coarser in spatial resolution (0.25 degrees) than the HCLIM simulations we have used, and Svalbard is an area of very steep topography, then the usefulness of the coarser scale simulations is drastically reduced and we have therefore not considered them for this study. This need for finer scale resolution is already evident in our results which show that variations in ROS and its characteristics can vary greatly with topography and thus such variations would be missed in the 0.25 degree CORDEX simulations. However, we do now have access to a total of 8 upcoming 0.11 degree Polar CORDEX simulations (from 4 RCMs and 2 GCMs) which have not yet been published on ESGF, so we intend to make an assessment of these simulations to improve the robustness of the study. Moreover, these simulations use the SSP3-7.0 scenario, which will also broaden our analysis.
For the trend detection, we have implemented a linear regression, using the slope of the linear model to give the annual trend and converted this to a decadal trend by multiplying the annual trend by ten. In our revisions we will include a statement outlining how the trend was calculated to inform the readers, and also illustrate time series of the four ROS characteristics from Longyearbyen, Hornsund and Ny Ålesund to show the variations in ROS characteristics representing contrasting inland and coastal climates as well as latitudinal variations.
With regards to the choice of time window (Nov-Apr) chosen for detecting winter ROS: the November to April period was originally chosen based on an earlier study where Synthetic Aperture Radar (SAR) wet snow detection had been used to detect ROS (Vickers et al., 2022) and was subsequently evaluated against CARRA (Vickers et al., 2024). This period was necessary because snow cover is typically transient in October and May is a month when spring snowmelt begins at low elevations, so the SAR approach is less useful to detect ROS in these months. However, we have now extended the analysis period from October to May (since the reanalysis dataset is not subject to the same limitations as SAR) and will update all figures based on the new period.
"l258: The future trends feature a maximum in Nov., a minimum in March and another maximum in April. This seems to be counter-intuitive and would deserve some discussion, or ideally an explanation."
- We expect further warming to lead to higher probability of ROS at the tail ends of winter when the average air temperature may be warm enough that rain falls instead of snow. Moreover, warmer conditions in the future in November will likely lead to later onset of snow cover in low lying coastal areas (i.e. November may become the new October with a transient snow cover, therefore fewer ROS days will be detected if no snow cover is present). In a warmer climate there will be a shift to more frequent ROS at higher elevations but potentially lower at the coast where snow cover may appear later than in the present climate. We intend to include in the revised version a discussion regarding April trends that warming could lead to higher probability of rain and/or potentially earlier spring snowmelt i.e. April could be the new May so ROS would be rain falling on a wet snowpack with less chance of ice formation (already mentioned)
Minor and editorial comments:
The introduction features some repetitions (e.g., the fact that ROS impact ecosystems) and would gain from streamlining/shortening.
We will review the introduction and exclude repetitions where relevant
Citation: https://doi.org/10.5194/egusphere-2025-2099-AC1
-
AC1: 'Reply on RC1', Hannah Vickers, 14 Sep 2025
-
RC2: 'Comment on egusphere-2025-2099', Anonymous Referee #2, 08 Aug 2025
General
This is a useful paper examining recent and future changes in rain on snow (ROS) events and their characteristics over Svalbard. ROS events in the Arctic have generated much attention in recent years due to their many impacts. This paper is a solid contribution. I have only a few comments and suggestions.
Specific
Abstract and elsewhere: It is widely viewed that SSPS-8.5 is an overly aggressive scenario. This needs to at last be discussed in the text.
Line 11: It is “specifically” duration, intensity seasonal timing, yes? (not “such as”)
Line 16: Following from the above comment, by “all characteristics” do you mean frequency, duration, and intensity? Please be specific.
With respect to CARRA, why only go back to 1991? Was it not run for earlier years?
Line 28: It is important to note that “Arctic Amplification” is very seasonal - largest in autumn and winter, and smallest in summer.
Line 91: 121.8 mm is extremely dry. For the reader not familiar with Arctic precipitation, it would be useful to point this out.
Section 3 and elsewhere: A pet peeve of mine: trends are positive or negative, not increasing or decreasing. In increasing trend implies that the trend is getting larger.
A question on Figure 3: Is part of the reason what there are no trends in April because over at parts of Svalbard, there is no snow on the ground? I think the answer is no, but a reader may be wondering. A clarifying sentence is warranted.
With respect to Figure 4, the paper could benefit from more discussion of the causes of precipitation changes. As discussed later in the paper, proximity to sea ice may play a role, but what about the idea that in warming climate the atmospheric carries more water vapor? Furthermore, there ae negative trends along the west coast, which argues for changes in circulation patterns. Also, what are the projected changes in sea ice? I assume that the ice margin by 2050-2070 has retreated well to the north.
Citation: https://doi.org/10.5194/egusphere-2025-2099-RC2 -
AC2: 'Reply on RC2', Hannah Vickers, 14 Sep 2025
Thank you for the constructive review and points raised. We hope the comments (italic) have been appropriately addressed below.
Abstract and elsewhere: It is widely viewed that SSPS-8.5 is an overly aggressive scenario. This needs to at last be discussed in the text.
Yes. We agree that SSP5-8.5 is a more extreme scenario based on current energy trends and socio-economic circumstances. Nonetheless, it is a plausible scenario and provides an upper bound on the range of possibilities. As such it highlights the full range of risks for policymakers, so that they can understand what happens if mitigation measures fail. We will also explore the possibility of using a set of 8 simulations, currently unpublished in ESGF, which have been produced at a coarser resolution (12-15km) and use SSP3-7.0 to broaden our range of future ROS scenarios for Svalbard.
Line 11: It is “specifically” duration, intensity seasonal timing, yes? (not “such as”)
Yes, and we will change the wording to “specifically”
Line 16: Following from the above comment, by “all characteristics” do you mean frequency, duration, and intensity? Please be specific.
Yes, we will emphasise specifically the characteristics referred to.
With respect to CARRA, why only go back to 1991? Was it not run for earlier years?
No, the present version of CARRA is only available from 1991 (data from 1990 was also available but not for a full year).
Line 28: It is important to note that “Arctic Amplification” is very seasonal - largest in autumn and winter, and smallest in summer.
We will emphasise this point in the revised version
Line 91: 121.8 mm is extremely dry. For the reader not familiar with Arctic precipitation, it would be useful to point this out.
We will include this detail in the revised manuscript
Section 3 and elsewhere: A pet peeve of mine: trends are positive or negative, not increasing or decreasing. In increasing trend implies that the trend is getting larger.
We will change instances of increasing/decreasing to positive/negative trends.
A question on Figure 3: Is part of the reason what there are no trends in April because over at parts of Svalbard, there is no snow on the ground? I think the answer is no, but a reader may be wondering. A clarifying sentence is warranted.
There are trends in all months, but we have chosen to only show where the trend was at least marginally significant i.e we have masked out areas with confidence > 0.1. So the lack of trend values in Fig.3 doesn’t mean no trends, it just means they were too weakly significant to be included. In the revised version of the manuscript we intend to show the trends for a wider time period (October to May) and only where statistically significant (p<0.05).
With respect to Figure 4, the paper could benefit from more discussion of the causes of precipitation changes. As discussed later in the paper, proximity to sea ice may play a role, but what about the idea that in warming climate the atmospheric carries more water vapor? Furthermore, there ae negative trends along the west coast, which argues for changes in circulation patterns. Also, what are the projected changes in sea ice? I assume that the ice margin by 2050-2070 has retreated well to the north.
We have currently argued that the negative trends close to the coast are more likely the result of shorter snow cover duration i.e. later onset, earlier snowmelt -> shorter snow cover = reduced time period when ROS can occur (lines 324-325). This scenario is also supported by the results published in a recent report by Landgren et al. (2025) that show there will considerable decreases in the fraction of spring and autumn precipitation falling as snow in the future scenario along the west and southern areas compared to the present day (this was not included but will be in the revised manuscript). However we agree that the point about the atmosphere carrying more water vapour in a warmer climate has been missed from the discussion (eg. Dobler et al., 2020) and should be included when considering these results. This will be included in our revised version. With regards to the sea ice margin we have included in the attachment an illustration of the margin for the 2000-2020 (present day), 2030-2050 and 2050-2070 periods. This shows that while the sea ice margin has indeed retreated further north in the western part of Svalbard, the most considerable changes can be observed in the eastern part of Svalbard.
-
AC2: 'Reply on RC2', Hannah Vickers, 14 Sep 2025
Viewed
HTML | XML | Total | BibTeX | EndNote | |
---|---|---|---|---|---|
662 | 62 | 18 | 742 | 14 | 27 |
- HTML: 662
- PDF: 62
- XML: 18
- Total: 742
- BibTeX: 14
- EndNote: 27
Viewed (geographical distribution)
Country | # | Views | % |
---|
Total: | 0 |
HTML: | 0 |
PDF: | 0 |
XML: | 0 |
- 1
The manuscript describes a trend analysis study of rain-on-snow (ROS) events in Svalbard. The authors use reanalysis data (CARRA, ERA5) and one dynamically downscaled climate simulation (HCLIM-MPI) for this purpose and found, that ROS events have been increasing in frequency, duration, and intensity in several regions in the past, and are expected to further increase in the future. To motivate this study, the authors provide a brief discussion of the ecological and hydrological consequences of ROS events and propose directions for future research.
Major comments:
A convection permitting RCM (HCLIM) is used for downscaling the single GCM used in this study. The authors argue, that this would lead to superior results compared coarser-scale RCMs (e.g., l74, l363), but miss to demonstrate this claimed added value. One reference is given (Landgren et al., 2025), but this is only a conference abstract with no information content (no results in the abstract). I.e., a basic evaluation and demonstration of added value of the HCLIM is completely missing, which leads to several complications. One of them is, that if no significant added value compared to other RCMs (e.g., from CORDEX-ARC-22) can be demonstrated, the authors could use an ensemble of conventional climate models instead. This would resolve the major weakness of this study (see following comment). Another issue of missing model evaluation is that the authors have to speculate on the reasons for differences between CARRA and HCLIM ROS characteristics. E.g., l312: “The discrepancy in absolute values of the characteristics for the present climate (2000-2020) are likely attributable to the uncertainty in temperature threshold for partitioning rain and snow used in the different datasets”. Couldn’t it also be a simple temperature bias in HCLIM?
My main concern about this manuscript is the use of only one GCM/RCM combination for the analysis of future trends. Particularly, the choice of the GCM can be expected to have large impact on the results. This weakness is clearly identified by the authors (l364), which is good, but at the same time, this is no justification, since other options would be available. E.g., the CORDEX-ARC-22 archive contains 79 simulations (https://esgf-node.ipsl.upmc.fr/search/cordex-ipsl/). The authors have to demonstrate how this single realization be regarded as representative, or at least show whether is a cool/warm and dry/moist realization of expected future climate. Or, preferably, use a comprehensive model ensemble, as it is state-of-the-art.
The method of trend detection is not named nor described. Please give a clear explanation of the statistical method used. This is particularly important, since the analyzed period (1991 – 2023) is rather short for detecting significant trend in time series with large variability. I additionally suggest presenting some time series, in order to allow the reader to get an impression of the variability involved.
l197: the results show a maximum of the trends in November. This would imply, that significant trends could also be present in October, or even earlier in the year. Please expand your evaluation time-window (currently Nov-April) accordingly, in order not to miss significant results.
l258: The future trends feature a maximum in Nov., a minimum in March and another maximum in April. This seems to be counter-intuitive and would deserve some discussion, or ideally an explanation.
Minor and editorial comments:
The introduction features some repetitions (e.g., the fact that ROS impact ecosystems) and would gain from streamlining/shortening.