An enhancement to sea ice motion and age products at the National Snow and Ice Data Center (NSIDC)

A new version of sea ice motion and age products includes several significant upgrades in processing, corrects known issues with the previous version, and updates the time series through 2018, with regular updates planned for the future. First, we provide a history of these NASA products distributed at the National Snow and Ice Data Center. Then we discuss the improvements to the algorithms, provide validation results for the new (Version 4) and older versions, and intercompare the two. While Version 4 algorithm changes were significant, the impact on the products is relatively minor, particularly for more recent years. The changes in Version 4 reduce motion biases by ∼ 0.01 to 0.02 cm s−1 and error standard deviations by ∼ 0.3 cm s−1. Overall, ice speed increased in Version 4 over Version 3 by 0.5 to 2.0 cm s−1 over most of the time series. Version 4 shows a higher positive trend for the Arctic of 0.21 cm s−1 per decade compared to 0.13 cm s−1 per decade for Version 3. The new version of ice age estimates indicates more older ice than Version 3, especially earlier in the record, but similar trends toward less multiyear ice. Changes in sea ice motion and age derived from the product show a significant shift in the Arctic ice cover, from a pack with a high concentration of older ice to a sea ice cover dominated by first-year ice, which is more susceptible to summer melt. We also observe an increase in the speed of the ice over the time series ≥ 30 years, which has been shown in other studies and is anticipated with the annual decrease in sea ice extent.


S1. Evaluation of uncertainties in source data motion estimates 1 2
The number and types of individual sources have changed over the ice motion product time series. 3 Each source has different error characteristics based on the source and the method to estimate 4 motion. As noted in the main text, Section 2.2, there have been several studies of motion estimate 5 uncertainties, most focused on passive microwave motions derived from SSMI/SSMIS and 6 AMSR-E imagery. These evaluations have been done via comparisons with buoy estimates. Buoy 7 position is known very precisely and thus displacement and motion can be accurately retrieved. 8 Thus, buoys act as a source of the "true" motion for validation. 9 10 Such comparisons were done early on in the product development and were used to derive the 11 relative quality of the different sourcei.e., the values of C in Equation 1. These C values have 12 been used since without change. Here we provide a brief analysis of the source data via comparison 13 with buoy source. This is not intended to be a complete validation study of each of the source 14 motions. It is meant to give users of the product a general sense of the accuracy of the source 15 motions. 16

17
The comparison method used all valid buoy motions over the given time period. For each buoy, 18 the closest source motion estimate was found with a limit of 50 km (only source observations 19 within 50 km of the buoy) were included. The u-component and v-component of motion, on the 20 EASE grid, were compared between the source and the buoy for each day. All comparisons over 21 the time period were used to calculate a bias (mean difference, e.g., usourceubuoy) and an Error 22 Standard Deviation (Table S1). For each source, a typical year was chosen and statistics were 23 calculated for winter (January through March) and summer (June through August). Particularly 24 for the microwave estimates, this represents a period of optimal performance (winter) and a period 25 with low performance (summer). For SMMR, summer estimates were very limited and there was 26 not a large enough sample to effectively calculate statistics, so only winter is provided. 27 28 Overall, the statistics recapitulate what has been shown in previous studies. Passive microwave 29 estimates generally have higher errors than AVHRR because of the lower spatial resolution. And 30 spatial resolution of the source passive microwave imagery is a key factor in the error; higher 31 spatial resolution imagery have lower Error Standard Deviation values. Note that SMMR Error 32 Standard Deviation values are relatively low because SMMR imagery is every other day and the 33 daily value is a result of averaging between days, which reduces the noise in motion estimates. 34

35
One feature that is noticeable is the high errors in SSMI during summer, particularly the bias. This 36 is not surprising given the difficulties in retrieving surface properties during the melt season. This 37 perhaps argues against inclusion of SSMI (and SSMIS, which has similar characteristics) in the 38 combined product during summer. However, the approach used in the product is to use all available 39 inputs throughout the year. During summer, the number of vectors is lower, particularly for 85 40 GHz, so the impact of these lower quality vectors is relatively small compared to the buoys and 41 the complete field of wind-derived motions. AMSR-E is also affected by surface melt, but its 42 summer errors are not as high as SSMI, likely due to the higher spatial resolution. 43

44
Of note also is that the wind-derived motion error statistics are comparable to the passive 45 microwave estimates, even lower in many cases (especially summer). This indicates that winds 46 should not necessarily be lower weighted than the other satellite sources. It also suggests that the 47 product's use of an ice/wind speed ration of 1% is not unreasonable, though 2% may yield 48 improvement, as shown in Table 4  The combined daily motions are validated in the Arctic through comparison with independent 56 buoys from the CRREL Ice Mass Balance Buoy program [Perovich et al., 2020]. We compared 57    Table S2. 80

81
As noted in Section 2.1 of the main text on Reanalysis winds, for the wind forcing, we used a 1% 82 scale factor for the ice speed relative to wind speed. Other assessments have shown that 2% may 83 be more legitimate, especially in recent years with the observed positive trend in ice speed. To 84 investigate the potential effect of underestimating ice speed from winds in our product, we 85 compared the combined motion fields with both 1% and 2% scale factors to the 2015 CRREL buoy 86 observations. There were a total of 2025 comparison for 2015. The results indicate little effect due 87 to wind speed (Table S3), which is expected since the weighting of the wind-driven motion is 88 lower than other sources. The comparison indicates that the magnitude of the bias changes little 89 for the u-component, but actually increases for the v-component. The error standard deviations 90 decrease, generally by ~0.5 cm/s. This suggests that 2% may indeed be an improvement, but the 91 impact on the combined gridded is relatively small. Of course, the relationship between wind speed 92 and ice motion is complicated and can be quite variable. It depends on the compactness of the ice 93 cover, thickness, and wind direction relative to nearby coasts. We plan to investigate the relationship further in the future, both regionally (for different sea ice conditions) and temporally 95 (to investigate the effect of the long-term trend toward increasing speeds). 96 97  Table S1. Error estimates for selected periods from the source motion estimates. All values are in 98 cm/s. The Number column indicates the total number of buoy-source matches used in the statistics. 99 The bias columns given the average difference in the motion components (source minus buoy) and 100 the SD diff column provides the standard deviation of the difference. 101 Year