the Creative Commons Attribution 4.0 License.
the Creative Commons Attribution 4.0 License.
| 13 May 2026
Changes in Arctic sea ice drift speed over the last 130 years
Abstract. The Arctic transpolar sea ice drift speed shows a pronounced increase over past decades, one of many manifestations of Arctic climate change. However, little is known so far how the drift changed in earlier periods. Here I use data from historical drift expeditions, in particular from the Fram in 1893–96, the Sedov in 1937–40, the Tara in 2007–08, and the Polarstern during MOSAiC in 2019–20, as well as from the Soviet/Russian North Pole drift stations, to derive a 130-year record of Arctic transpolar drift speed. The transpolar drift speed already increased significantly during the early 20th century warming, followed by a period of slowing drift in the 1950s–70s and a strong increase in recent decades, closely following the evolution of Arctic mean temperatures. The observed fractal scaling of the drifts can be explained quantitatively by a Brownian motion random walk process that includes temporal auto-correlation and a mean drift term due to currents and prevailing winds. Comparisons of the sea ice drift speeds with near surface wind observations reveal that the long-term changes in drift speed are not primarily caused by changes in wind speed.
- Preprint
(2139 KB) - Metadata XML
- BibTeX
- EndNote
Status: final response (author comments only)
-
RC1: 'Comment on egusphere-2026-2561', Dmitry Divine, 14 Jun 2026
-
AC1: 'Reply on drift along east coast of Greenland', Bjoern-Martin Sinnhuber, 23 Jun 2026
The comment was uploaded in the form of a supplement: https://egusphere.copernicus.org/preprints/2026/egusphere-2026-2561/egusphere-2026-2561-AC1-supplement.pdf
-
AC3: 'Reply on RC1', Bjoern-Martin Sinnhuber, 13 Jul 2026
Sincere thanks to reviewer Dmitry Divine for the detailed and constructive comments on the manuscript. Below I give a point-by-point response to his comments. I already published a first response (https://doi.org/10.5194/egusphere-2026-2561-AC1) to present an additional analysis of buoy data to better constrain the drift along the east coast of Greenland.
The paper is well written in my opinion, and I got only one moderate to major comment. I believe the author can elaborate more on the issue he raised at the very end of the Discussion part, namely the drift of the wreckage of Jeannette to south-western Greenland. The wreck was found drifting with the ice, not stranded ashore. This provides quite reliable starting point for some basic analysis of the timing this could take for the wreckage to drift along the eastern shore of Greenland towards the recovery location. With these estimates at hand, one can have a better idea of how long it took for the wreckage to drift across the central Arctic, hence adding another valuable point in Figure 5. The author can use the data from https://iabp.apl.uw.edu to identify the buoys that travelled all the way to south-west Greenland with the E(W)GC and making drift duration estimates. Even if the values are to be very likely biased low due to the observed recent acceleration of the TPD, those will undoubtedly resolve if this could last “at least 400 days” as was postulated by Nansen back in the days.
Thank you for the suggestion to look into buoy data to provide more constraints on the drift along the east coast of Greenland and with this to better constrain the drift of the Jeannette wreckage across the Arctic Ocean. A detailed response was given in my earlier comment https://doi.org/10.5194/egusphere-2026-2561-AC1: While a comprehensive analysis of all available buoy data contains more than enough material for an own publication beyond the scope of the present study, I decided to illustrate the drift along the east coast of Greenland with the example of one particular buoy. Interestingly, this buoy drifted in 2022-23 at essentially the same speed as the Soviet North Pole station NP-01 in 1937-38. I believe this example provides an interesting new aspect, worth to be included in the revised version of my manuscript with an additional Figure 10. However, I do not believe that it is possible to constrain this enough to take the drift of the Jeannette as another independent data point for the transpolar drift. Results are shown in AC1.
Minor comments.
Line 120: …”but not considered in the present study”. Why these observations were discarded?
One chief difficulty comes from the fact that winds are measured at different heights for the different drift expeditions. A quantitative analysis thus requires further in-depth analyses of the wind forcing beyond the scope of the present study.
l.118-120 now changed to “Near-surface wind observations of NP-01 through NP-31 are published as part of the data set by Fetterer and Radionov (2000). Observations have been performed at different heights above surface: 2 m (NP-01), 6 m (NP-10, NP-13, NP-15), 8 m (NP-04, NP-06), and 10 m (NP-22 – NP-28).” Similarly for MOSAiC l. 105-107 changed to: “Meteorological observations were performed during MOSAiC from various platforms on the ship and on the ice (Shupe et al., 2022). Winds were measured from Polarstern at 37 m above surface and from a 10-m tower on the ice at 2, 6, and 10 m heights (Cox et al., 2023).” In the Results section following l. 171 the following statement will be included: “Wind observations from the North Pole drift stations likewise show no long-term trend. While for all drifts there is a strong correlation between wind speeds and drift speeds at the daily to monthly scale, the fact that wind observations for different expeditions are taken at different heights makes it difficult to conclude on changes in the wind factor on inter-annual to decadal time scales.”
Line 145: “the absolute amounts of the drift speeds” – may be “values”?
Okay, changed
Line 205: “…drift speeds as a function of the time intervals between the two positions”. How was this calculated? Random draws from the data set?
Thanks for comment, this was indeed not clear. I will revise/expand the description at the beginning of Section 2. For interval lengths of n days, the results were averaged over the n different realizations, each one day apart. Similarly, the analysis shown in Fig. 7 uses hourly MOSAiC data, where for interval lengths of n hours the results were averaged over the n different realizations.
Figure 6: “…indicate the 2-sigma uncertainty derived from the observed drift speeds”. Is this sigma corrected for autocorrelation?
Yes. The errorbars in Fig. 6 were calculated consistent with the description in the text (l. 177-179): “Errorbars indicate the 2σ-standard error of the mean, calculated by the standard deviation of the daily drift speeds, divided by the square root of the effective degrees of freedom 𝑛eff = 𝑛(1 − 𝑟)/(1 + 𝑟), with n the number of days and 𝑟 ≈ 0.6 the 1-day lag auto-correlation coefficient of the time series.”
Figure 8: I recommend choosing different color scheme or/and use different symbols, otherwise hard to distinguish between different circles. Also, daily ice drifts are shown for mixed wind speeds for the three different experiments; I would suggest applying correcting factors to put them on the same wind speed scale, say 10m, even if it leads to some additional uncertainty.
Okay, I will try to improve the colour scheme and or use of different symbols in Fig. 8. I will follow the suggestion to scale the observed wind speeds to a common reference of 10 m, using the relation from Tara.
Line 247: “…and concentration, leading to a faster drift…” Changing sea ice roughness/oceanic and atmospheric drags also considered as one of the factors (see e.g. Krumpen, T., et al. Smoother sea ice with fewer pressure ridges in a more dynamic Arctic. Nat. Clim. Chang. 15, 66–72 (2025). https://doi.org/10.1038/s41558-024-02199-5) though its significance in the overall picture is not yet fully understood.
Thanks for the comment and for pointing out to the paper by Kumpen et al. (2025). I will include a reference to Krumpen et al. (2025) and include this in the discussion (see following point).
Eq.(1) is it valid for free drift? Note that when making estimates below, the changes in the drag coefficients can also be considered.
Yes, Eq. (1) includes the effects of atmospheric and ocean drag, Coriolis force and sea surface tilt. But it does not include internal ice stress, so is valid only for free drift. I am grateful for the comment on changes in drag coefficients and for pointing to the study by Krumpen et al. (2025). I will thus include the following in the discussion: “Krumpen et al. (2025) showed that Arctic sea ice has become smoother over the last decades, resulting in a reduction of the atmospheric drag coefficient. However, a reduction in the atmospheric drag coefficient would lead to a slower sea ice drift, not a faster drift as observed. It seems reasonable to assume that the ice not only becomes smoother on the upper side, but at the same time also on the bottom side, with a corresponding reduction of the water drag coefficient as well, although Krumpen et al. (2025) could not show that. Eq. (1) reveals that if the atmospheric and water drag coefficients are reduced simultaneously by the same factor, this is mathematically equivalent to an increase in ice thickness and will thus also result in a slower drift, not a faster drift.”
Line 278: “…by scaling the random speeds by 1/sqrt(((1-r)/(1+r)))…” I assume this is to account for the increasing effect of autocorrelation at shorter time intervals; should the sqrt(n) factor be also included? Or it applies to daily velocities only?
Maybe writing “…by scaling the random forcing…” (instead of “random speeds”) would make this point clearer. The inclusion of autocorrelation without rescaling leads to a reduced variance at daily intervals, that is not in agreement with the observed variance (such as that shown in Fig. 3) anymore. This is independent of the number of days n, so no scaling of sqrt(n) to be included.
Line 279:”…resulting then in an increased variance at long intervals (cyan lines in Figure 9).” The variance would increase for the ordinary Brownian motion too (see black lines in Figure 6), but positive autocorrelation would make the process to spread faster than a standard random walk by a factor of (1+r)/(1-r).
Thanks. Will be reformulated to “…resulting then in an increased drift speed at long intervals (cyan lines in Figure 9)“ as this is in fact the relevant point to be made here.
Line 279: ” …inclusion of a mean drift term results….” I believe adding another equation (e.g. into Appendix) for a modified Brownian motion could be a good idea.
Okay, I will include an equation to make this clearer.
Line 301 “…to the detection of the remnants” – may be “finding/recover of the wreckage of” can be a better formulation?
Okay, will be changed accordingly.
Line 314 ”…provide a 130-years time series of Arctic sea ice drift…” I believe this statement is far too ambitious given the scarcity of the data in the early part of the period considered. Please reformulate.
Thanks, I agree. Will reformulate as: “The historical drift expeditions provide information on Arctic sea ice drift speed extending back in time and that agrees well with estimates of the increase in Arctic sea ice drift speeds over recent decades (e.g., Rampal et al., 2009).”
Similarly, the abstract l. 8-9 will be changed to “…to provide information on the Arctic transpolar drift speed over the last 130 years.”
Citation: https://doi.org/10.5194/egusphere-2026-2561-AC3
-
AC1: 'Reply on drift along east coast of Greenland', Bjoern-Martin Sinnhuber, 23 Jun 2026
-
RC2: 'Comment on egusphere-2026-2561', Anonymous Referee #2, 01 Jul 2026
This paper studies changes in drift speed using data from the Fram, Sedov, Northpole Stations and the recent drifts of the Tara and MOSAIC expeditions to produce a “130 year” record of drift, and they further study differences in the drift speed over time. They find that there are changes in drift speed and find a correlation between windspeed and temperature. They also find brownian and fractal qualities to the motion of sea ice.
My primary concern with this paper is What is new and justifies publication?
- Their finding that drift speed is associated with temperature and thinner ice has been known for decades.
- Looking at Figure 5 and Table 1, I see no trend in the drift speed over the 1890 to 2000 period, then a remarkable difference in speed of the Tara and MOSAIc cruises. This has been documented in papers they cite: Krumpen, et al. 2021; Kwok et al. 2013; and Vihma et al., 2008.
- Studies of the brownian motion, spatial and temporal scaling of sea ice motion has also been published, e.g. Weiss 2013, and many papers by Thorndike and Colony back in the 1980s.
Citation: https://doi.org/10.5194/egusphere-2026-2561-RC2 -
AC2: 'Reply on RC2', Bjoern-Martin Sinnhuber, 01 Jul 2026
I thank Referee 2 for the feedback on my manuscript. The Referee’s primary concern with this paper is What is new and justifies publication? So, I start my reply by a list of novel points in this study, that to the best of my knowledge have not been shown before. This is followed by a point-by-point response to the specfic Referee's comments.
- For me the most important point is the finding that the Arctic sea ice drift has accelerated already during the early 20th century warming period. There is very little information of Arctic sea ice changes during the early 20th century warming period and very few observations of changes and impact in the central Arctic in general. The drift of the Sedov, following more or less the path of the Fram drift, but at the peak of the early 20th century warming period, provides an ideal opportunity to investigates this. Of course, the drift of the Sedov has been discussed in the scientific literature before, as early as in Zubov’s 1940 Nature paper cited in my manuscript. But to the best of my knowledge, there is no previous publication that analyses in detail the drift speed of the Sedov in comparison with the Fram and more modern drifts.
- My analysis shows a statistically significant increase of the transpolar drift speed between the Fram and the Sedov at the 2-sigma level. The comment “Looking at Figure 5 and Table 1, I see no trend in the drift speed over the 1890 to 2000 period, then a remarkable difference in speed of the Tara and MOSAiC cruises” in my view misses the point. It is not about possible long-term trends independent of the Arctic warming, but investigating how the drift speed changes with changes in Arctic temperatures over previous times. The data in Table 1 show a correlation between drift speed and Arctic-wide temperature anomalies of r=0.95 for the whole period, and still a correlation coefficient of r=0.67 (statistically significant at the 95% confidence level for a two-sided Student t-test) when excluding the Tara and MOSAiC drifts after 2000. To the best of my knowledge, an analysis of these changes (including their statical significance) for the almost 100 years between the Fram drift and the 1980s has not been published before.
- The investigation of the statistical distribution of measured wind speeds and the relation between wind and sea ice drift speed shows clearly that the (statistically significant) increases in drift speeds between Fram, Sedov, and Tara are not due to changes in wind speeds. While previous studies have already provided evidence that changes in wind speeds are not the primary drivers of the increase in Arctic sea ice drift speed, I am not aware that this has been shown explicitly before for these three historical drift expeditions. (Vihma et al., 2008, discuss only in general terms that the winds during the Tara drift were relatively low, but don’t compare this quantitatively with earlier drifts.)
- The scaling of the drift speeds with the interval length as shown in my Figs. 6 and 7 demonstrate that the drifts increased at all interval lengths. This provides clear evidence that the acceleration of the drifts is not (or not primarily) due to changes in ocean currents: these would have affected only the mean drift at long intervals and would have only a small impact on the drifts measured at daily intervals. Of course, it has been shown before that the drift speed changes with the interval considered. Frolov et al. (2005) have provided an empirical scaling of the drift speed that agrees well with my analyses and is shown for comparison in Fig.6. But this empirical relation is valid only for a restricted range of intervals. I am not aware of any previous publication that has shown similar figures as in my Fig. 6 and how the scaling changes for short and long intervals.
- We can understand the scaling of the drift speeds with interval lengths from very simple principles, as illustrated in my Fig. 9. This provides also a quantitative explanation of the empirical scaling as given by Frolov et al. (2005). And as importantly, it provides a quantitative explanation of where this single scaling is not applicable any more. I am not aware of any previous study that has shown this. In this respect, I am grateful to Reviewer 2 for the reference to the book by J. Weiss (2013) that I did not know before and have now ordered.
- The scaling, as shown in my Fig. 6 and schematically in Fig. 9 can be quantitatively understood from the basic properties of the underlying Rice distribution, as summarized in Appendix A of my manuscript. The Rice distribution is in my view a rather natural assumption for wind speeds and drift speeds, as it is the theoretical distribution to be expected when the components of the wind or drift are Gaussian distributed. I am not aware that this has been discussed in the sea ice literature before, and it has only very recently received attention (independently from my study) in the meteorological literature.
- In general terms it has long been suggested that the observed increase in sea ice drift speed may be related to a reduction in thickness, strength and concentration of sea ice. And while the direct relation between sea ice thickness and drift speed under the assumption of free drift has already been shown by Hibler (1986) and reproduced by my Eq. (1) and Fig. 8, the implications of this have received little attention so far. While it is clear that Eq. (1) neglects the internal ice stresses, it nevertheless provides a useful theoretical concept to discuss the relation between wind forcing and sea ice drift as a function of ice thickness. With all these limitations, isn’t it remarkable how well Eq. (1) reproduces the observed relation between wind speed and sea ice drift for the Tara expedition, even quantitatively when using typical and well accepted assumptions for the drag coefficients (Fig. 8)? In this respect I am also grateful to Referee 1 for pointing to the study by Krumpen et al. (2025) who showed that Arctic sea ice has become smoother over the last decades. Krumpen et al. (2025) find that this results in a corresponding reduction in the atmospheric drag coefficient, which however would result in a slower sea ice drift, not a faster drift as observed. It seems reasonable to assume that the ice not only becomes smoother on the upper side, but at the same time also on the bottom side, with a corresponding reduction of the water drag coefficient, although Krumpen et al. (2025) could not show that. However, Eq. (1) reveals that if the atmospheric and water drag coefficients are reduced simultaneously by the same factor, this is mathematically equivalent to an increase in ice thickness and will thus result in a slower drift, not a faster drift! I am not aware that this has been discussed anywhere in the literature so far.
I will try to revise the conclusions of my manuscript to make these points clearer.
Specific points:
“1. Their finding that drift speed is associated with temperature and thinner ice has been known for decades.”
I agree that it is well accepted that the increase in drift speed is related to thinner ice in a warmer Arctic. However, I am not sure if there is a consensus as to the relative importance of factors like sea ice thickness, strength, or concentration, the importance of changes in drag coefficients, changes in ocean currents, or even wind forcing. In my manuscript I have accordingly tried to cite much of the relevant previous works and tried to discuss where the findings of the present study confirm previous results and where they challenge previous claims.
“2. Looking at Figure 5 and Table 1, I see no trend in the drift speed over the 1890 to 2000 period, then a remarkable difference in speed of the Tara and MOSAiC cruises. This has been documented in papers they cite: Krumpen, et al. 2021; Kwok et al. 2013; and Vihma et al., 2008.”
See my comment above: the question is not whether or not there is a trend over the 1890 to 2000 period, but in my view the important finding is that the Arctic transpolar drift speed increased during the early 20th century warming period, followed by a slight slow down during the 1950s-1970s, in line with changes in Arctic-wide temperatures. The change in drift speed between Fram and the Sedov is significant at the 2-sigma level and Fig. 6 shows that the change occurred at all scales. The correlation between drift speed and Arctic-wide temperature anomalies for the period 1890 to 2000 (i.e. excluding the Tara and MOSAiC drifts) shows a correlation coefficient of r=0.67 which is statistically significant at the 95% confidence level. In hindsight, the difference in speeds for Tara and MOSAiC compared to earlier drifts is maybe not so “remarkable”, but well in line with the increased Arctic warming as shown in Fig. 5.
“3. Studies of the brownian motion, spatial and temporal scaling of sea ice motion has also been published, e.g. Weiss 2013, and many papers by Thorndike and Colony back in the 1980s.”
I am grateful to Referee 2 for pointing to the book by Weiss (2013) that I didn’t know before but have now ordered. I agree that many important concepts and results have already been presented by Thorndike and Colony in the 1980s, that I have already cited. In my view the important new result of the present study in this respect is to show how the scaling of the drift speed with the interval length changes with scales (i.e. the difference between a fractal and a multifractal scaling, although I don’t like the expression “multifractal” too much) and by providing a very simple yet quantitative explanation for this scaling behaviour.
Citation: https://doi.org/10.5194/egusphere-2026-2561-AC2
Viewed
| HTML | XML | Total | BibTeX | EndNote | |
|---|---|---|---|---|---|
| 208 | 55 | 14 | 277 | 18 | 13 |
- HTML: 208
- PDF: 55
- XML: 14
- Total: 277
- BibTeX: 18
- EndNote: 13
Viewed (geographical distribution)
| Country | # | Views | % |
|---|
| Total: | 0 |
| HTML: | 0 |
| PDF: | 0 |
| XML: | 0 |
- 1
Review of egusphere-2026-2561
Changes in Arctic sea ice drift speed over the last 130 years by Sinnhuber, B.-M.
Summary:
The manuscript analyzes observations from historical Arctic drift expeditions, including both ship-based expeditions and drifting stations established on sea ice, beginning with the Fram expedition. The author investigates changes in drift duration and velocity across the Arctic Basin and discusses these changes in relation to long-term and recent trends in Arctic climate and sea-ice cover. Analyzing the changes in drift velocities and wind factors through time, the author proposes air temperature-driven changes in ice thickness to be a major factor controlling the variations in average transpolar sea ice drifts speeds in the central Arctic Basin. The author also calculates the scaling exponent of drift velocities for historical drifts and demonstrates the similarity of the results with a random walk model with autocorrelation and a mean drift term. I prefer to refrain though from commenting on the implications of this part of the analysis, since I am not an expert in sea ice kinematics and dynamics at such level of detail.
The paper is well written in my opinion, and I got only one moderate to major comment. I believe the author can elaborate more on the issue he raised at the very end of the Discussion part, namely the drift of the wreckage of Jeannette to south-western Greenland. The wreck was found drifting with the ice, not stranded ashore. This provides quite reliable starting point for some basic analysis of the timing this could take for the wreckage to drift along the eastern shore of Greenland towards the recovery location. With these estimates at hand, one can have a better idea of how long it took for the wreckage to drift across the central Arctic, hence adding another valuable point in Figure 5. The author can use the data from https://iabp.apl.uw.edu to identify the buoys that travelled all the way to south-west Greenland with the E(W)GC and making drift duration estimates. Even if the values are to be very likely biased low due to the observed recent acceleration of the TPD, those will undoubtedly resolve if this could last “at least 400 days” as was postulated by Nansen back in the days.
Minor comments.
Line 120: …”but not considered in the present study”. Why these observations were discarded?
Line 145: “the absolute amounts of the drift speeds” – may be “values”?
Line 205: “…drift speeds as a function of the time intervals between the two positions”. How was this calculated? Random draws from the data set?
Figure 6: “…indicate the 2-sigma uncertainty derived from the observed drift speeds”. Is this sigma corrected for autocorrelation?
Figure 8: I recommend choosing different color scheme or/and use different symbols, otherwise hard to distinguish between different circles. Also, daily ice drifts are shown for mixed wind speeds for the three different experiments; I would suggest applying correcting factors to put them on the same wind speed scale, say 10m, even if it leads to some additional uncertainty.
Line 247: “…and concentration, leading to a faster drift…” Changing sea ice roughness/oceanic and atmospheric drags also considered as one of the factors (see e.g. Krumpen, T., et al. Smoother sea ice with fewer pressure ridges in a more dynamic Arctic. Nat. Clim. Chang. 15, 66–72 (2025). https://doi.org/10.1038/s41558-024-02199-5) though its significance in the overall picture is not yet fully understood.
Eq.(1) is it valid for free drift? Note that when making estimates below, the changes in the drag coefficients can also be considered.
Line 278: “…by scaling the random speeds by 1/sqrt(((1-r)/(1+r)))…” I assume this is to account for the increasing effect of autocorrelation at shorter time intervals; should the sqrt(n) factor be also included? Or it applies to daily velocities only?
Line 279:”…resulting then in an increased variance at long intervals (cyan lines in Figure 9).” The variance would increase for the ordinary Brownian motion too (see black lines in Figure 6), but positive autocorrelation would make the process to spread faster than a standard random walk by a factor of (1+r)/(1-r).
Line 279: ” …inclusion of a mean drift term results….” I believe adding another equation (e.g. into Appendix) for a modified Brownian motion could be a good idea.
Line 301 “…to the detection of the remnants” – may be “finding/recover of the wreckage of” can be a better formulation?
Line 314 ”…provide a 130-years time series of Arctic sea ice drift…” I believe this statement is far too ambitious given the scarcity of the data in the early part of the period considered. Please reformulate.