Essential gaps and uncertainties in the understanding of the roles and functions of Arctic sea ice

The sea ice concentration and extent record from passive microwave instruments provides a nearly complete and consistent record since late 1978. On one hand, this almost 40 year long time series is a very important tool for climate studies, and it is used widely as a climate indicator to quantify and illustrate Arctic sea ice change. On the other hand, the fact that the time series is limited to four decades, that it has limited spatial resolution, some varying data quality, and uncertainty depending on season and changing satellite sensors over time all limit the applicability of the time series. Intercalibration between sensors has not always been optimal (Eisenman et al 2014). This could be improved via use of longer overlaps period that are now available, employing better calibrated source data and, as is now being done in some products (Comiso et al 2017, Lavergne et al 2019), implementing adaptive time-varying algorithm coefficients that adjust for changing surface conditions. Nevertheless, there is a looming gap due to the potential loss of satellite coverage. Passive microwave sensors have been routinely launched since 1987 and for many years multiple sensors have been in orbit. However, recent failures have reduced the number currently operating sensors and have significantly increased the risk of a gap in coverage in the next years. AMSR-E failed in 2011, and the DMSP F-19 SSMIS failed in early 2016 after only two years of operation. In April 2016, the F-17 SSMIS started behaving anomalously and its data quality has been reduced. The suite of DMSP SSMI/SSMIS instruments has been the workhorse of passive microwave observations, but all currently operating (as of June 2018) SSMIS instruments (F-16, F-17, F-18) have been in orbit for at least 8.5 years, well beyond their design lifetime of three years (figure 1). The JAXA AMSR2 sensor, operating for a little over six years (as of June 2018) is the youngest passive microwave sensor, but it is now also past its design lifetime (five years). The only remaining sensor potentially ready to be launched in the near future (the DMSP F-20 SSMIS) has been cancelled.

Zoom In Zoom Out Reset image size

While discussions concerning follow-on sensors are ongoing (including an AMSR follow-on, a EUMETSAT instrument on their METOP constellation, and possible new US radiometers) these are all likely to be a minimum of five years from launch. At that point, the youngest sensor commonly used for climate monitoring (AMSR2), if still operating, would be nine years old. There is a passive microwave sensor (MWRI) on the Chinese FengYung-3B and -3C platforms. This has not been employed as part of the long-term passive microwave sea ice climate record, but has potential to fill the looming gap due to the loss of current capabilities. First intercalibration efforts of the FY-3C satellite data with other passive microwave satellite data have been made (Li et al 2016, Wang et al 2018, Wu and Liu 2018).

Another important, but more difficult to measure sea ice climate indicator is sea ice thickness. Significant development has been made to quantify sea-ice thickness and its seasonal and interannual change. However, there are still a number of significant gaps in knowledge about sea-ice thickness and the methods for measuring it. One method to estimate sea-ice thickness is to measure freeboard from aircraft or satellite and calculate ice thickness from that. Here, there is a gap in interpreting the radar and laser returns to yield accurate freeboard measurements (specifically for radar altimetry; see recent work by Ricker et al 2017, King et al 2018). Second, better snow and ice density information is needed to convert freeboard estimates into ice thickness. The most significant gap in this context, however, is the lack of reliable basin-scale snow depth observations (see e.g. Webster et al 2018). The snow weighs down the ice, changing the freeboard to thickness ratio. For laser altimeters, which reflect off the top of the snow surface, the information about snow depth is only required for the freeboard-to-thickness conversion. For radar altimeters, which penetrate the snow depending on its properties, the snow depth is required for both, accurate freeboard retrieval and freeboard-to-thickness conversion. The comparably poor spatiotemporal coverage with altimeter measurements and changing satellite sensors limit the present-day maturity of sea-ice thickness products further. Other methods that do not have the same problem are in situ surveys with classical thickness measurements from drillings, and those using electromagnetics (from air and ground), thermal methods (e.g. Mäkynen and Karvonen 2017) or microwave radiometry like from SMOS (e.g. Kaleschke et al 2016). However, these types of measurements have a number of limitations. Drillings are only possible in a few places of the Arctic at a given time. Electromagnetic measurements do not distinguish between snow and sea ice, and wrong information of snow thickness can lead to over- or underestimation of the ice thickness. Thermal measurements are only useful and provide accurate thickness information for relatively thin sea ice, i.e. below ∼0.5 m, and the same applies for the usage of microwave radiometry.

It is also important to mention that CryoSat-2 is beyond its life time, and that CryoSat-2 is the only currently flown radar altimeter providing data north of 81.5° N, covering the Arctic Ocean up to 88° N. The laser altimeter on ICESat-2, launched in September 2018, covers latitudes up to 86° N. In any case there remains an observation gap for the central Arctic Ocean.

The combination of ice extent, ice concentration and ice thickness information to ice volume estimates is another important climate indicator, and has been used in local (Gerland and Renner 2007) and panarctic contexts (Laxon et al 2013). Sea ice volume estimate is a powerful climate indicator, because it takes into account dynamic effects, which can lead to apparent reductions or increases of sea ice amounts due to for example ridging, rafting and spreading existing sea ice into larger areas. Complete sea ice volume datasets are rare since they require both ice extent, concentration and thickness data with sufficient spatial and temporal resolution (e.g. Kwok and Cunningham 2015, Lindsay and Schweiger 2015, Tilling et al 2018). The uncertainty of such sea-ice volume estimates naturally depends on the uncertainties of the input data which we discuss in the following sections. Alternatively, coupled ice-ocean models into which sea-ice information is assimilated, as e.g. PIOMAS (Zhang and Rothrock 2003), provide a reasonable measure of the Arctic sea-ice volume, which can be applied for further studies, such as on sea ice volume fluxes (e.g. Schweiger et al 2011, Zhang et al 2012, Zhang et al 2017). However, evaluating the quality of such results in the context of their value as an independent powerful climate indicator remains a challenge and requires careful evaluation taking into account the model's capability to properly resolve physical processes at the grid resolution used.

Estimates of uncertainty from passive microwave sea-ice concentration (SIC) products is a challenge owing to the large scale, varying surface properties, and limited availability of comparison data; summer is especially challenging due to the effect of melt water on the passive microwave signal (e.g. Kern et al 2016). Evaluation of SIC products is an ongoing effort—because new algorithms are still being developed (Kongoli et al 2011, Tikhonov et al 2015), because new satellite sensors are becoming available (e.g. AMSR2, Beitsch et al 2014), and because new independent data sets are emerging which help to better quantify SIC uncertainties. SIC derived from clear-sky satellite optical imagery such as MODIS for summer Arctic sea ice (Rösel et al 2012) can help quantify uncertainties in satellite microwave radiometry-based SIC during summer melt (Kern et al 2016), but such sensors depend on clear sky and daylight conditions. Thin ice thickness distribution derived, under freezing conditions and for close to 100% sea ice, from the ESA SMOS sensor (Kaleschke et al 2012, Huntemann et al 2014, Tian-Kunze et al 2014, Kaleschke et al 2016) can be used to infer biases in SIC over thin ice (Heygster et al 2014).

More effort has been dedicated during recent years to assess the retrieval uncertainty of SIC or at least providing SIC data sets with an uncertainty estimate—which is an important pre-requisite for assimilating SIC data into numerical models. A few approaches have been developed. One is to apply Gaussian error propagation to the equations used to estimate SIC and to calculate SIC uncertainty as a function of algorithm tie point, brightness temperature, and gridding uncertainties (Kern 2004, Spreen et al 2008). This is used in the EUMETSAT OSI-SAF SIC data set (Tonboe et al 2016, Lavergne et al 2019) and the ESA-CCI SIC data set (http://esa-cci.nersc.no, Lavergne et al in 2019). A second approach is to take the standard deviation of the SIC within a certain area as a measure for the quality of the SIC estimate (Peng et al 2013, Meier et al 2014) based on the fact that regions of higher SIC variability will have higher uncertainty because of the low sensor spatial resolution. The Enhanced NASA Team ('NASA Team 2') algorithm takes advantage of the iterative nature of the algorithm and estimates the uncertainty from the variability of the concentration as the iteration converges to a minimum cost function (Brucker et al 2014). These approaches provide a measure of the precision only.

More difficult is a quantitative estimate of the uncertainty of the sea-ice area or extent, which goes beyond a simple computation of the variation in these quantities over time. Factors contributing to uncertainty in sea-ice area and extent are (i) the choice of the threshold ice concentration used, (ii) filters used to flag spurious weather-influence induced SIC over open water (see e.g. Lavergne et al 2019), (iii) the accuracy of the SIC itself (e.g. low biases during melt conditions), (iv) filters used to flag land-spillover effects, and (v) use of different land masks.

Meier and Stewart (2019) presented an approach to quantify the variability in extent based on the above five factors as a proxy for an uncertainty estimate. They compared sea ice total extent from several different products and found a variability range of 500 000 square kilometers or more, depending on season. This generally corresponds to relative biases in ice edge location of 25–50 km and is due the factors mentioned above. The differences in extent, while varying seasonally, are consistent over the years for similar times of years. This means that trends and variability between the different products are more consistent than the absolute extents. In another part of the Meier and Stewart (2019) study, they found that using a consistent algorithm and processing yielded very stable results with standard deviations estimates at ∼20 000 square kilometers. However, estimating the accuracy of sea ice area remains challenging because of potential biases in the SIC due to snow and sea-ice surface processes and weather effects not accounted for in the algorithms.

Another aspect is the uncertainties in trends. Trend uncertainties are often given as the uncertainty (standard deviation) of the linear fit. However, trend uncertainties are affected by the limited consistency of the satellite record. This issue has not been thoroughly investigated, but Eisenman et al (2014) found that trend uncertainties may be higher than the quoted standard deviation of the trend. Another issue when examining trends and anomalies is the common assumption that while absolute extent or area values may be biased due to surface melt, snow, and ice conditions, confidence in trends would still be high because these surface effects are consistent over the years (i.e. every summer, surface melt causes an underestimation in concentration and area). This assumption, however, might not be valid anymore because of the observed earlier melt onset and the occurrence of more highly degraded or 'rotten' ice and resulting underestimation of the ice cover by passive microwave instruments (Barber et al 2009, Meier et al 2015). Specifically, for certain times of the year negative trends in sea-ice extent could appear larger than they are.

Regarding ice thickness from altimetry, Zygmuntowska et al (2014) reported that when uncertainties in the input parameters for the freeboard-to-thickness conversion are considered appropriately, the changes in sea-ice volume between the ICESat and CryoSat-2 periods are less dramatic than was reported in the literature at that time; this view was later shared by Kwok and Cunningham (2015). Uncertainty sources in the freeboard-to-thickness conversion and the freeboard retrieval itself are manifold. Various publications have shown that both sea ice and snow density play a major role in the uncertainty budget of the sea-ice thickness derived from satellite radar altimetry (e.g. Alexandrov et al 2010, Forsström et al 2011, Kern et al 2015).

The validation of these data is in its infancy. Many uncertainties still exist in the derivation of the basic parameter, the freeboard (Kwok 2014, Ricker et al 2015). In addition, data used for the validation might also not be free of potential biases or have unknown uncertainty (e.g. Kurtz et al 2013, Lindsay and Schweiger 2015), which complicates validation procedures.

Laxon et al (2013) used additional data to distinguish first-year and multi-year ice and applied different sea ice densities and snow depth values for the freeboard-to-thickness conversion. Zygmuntowska et al (2013) suggested an approach that would allow this discrimination by using CryoSat-2 data itself—without the need for additional data and with the potential of improved freeboard retrieval.

Several publications (Ricker et al 2014, 2017, Kurtz et al 2014, Tilling et al 2015, 2018) address enhanced solutions for the freeboard retrieval and freeboard-to-thickness conversion using CryoSat-2 data. The limited range resolution of CryoSat-2, the unknown tracking point of the impinging radar wave in the ice-snow system, the change in radar wave propagation speed in a snow layer of unknown thickness, and the presence of saline snow on seasonal sea ice can complicate the freeboard retrieval (Kwok 2014, Nandan et al 2017). One of the main sources of uncertainty is the snow depth on sea ice (Forsström et al 2011, Kwok 2014, Ricker et al 2015). Its retrieval remains a challenge.

The laser altimetry satellite ICESat ceased operation in 2009, but is still useful for extending the CryoSat-2 thickness time series backwards. However, stitching together these products is difficult considering the still limited knowledge regarding uncertainties. For example, Bi et al (2014) analyzed ICESat data and found large estimation errors over thick ice. Radar altimeter data were also provided by the ERS1/2 satellites (Laxon et al 2003) and the Envisat satellite (Giles et al 2008, Schwegmann et al 2016, Paul et al 2018). One of the challenges here, which is only solved partly (e.g. Paul et al 2018), is to produce a consistent data record of freeboard and sea-ice thickness extending from the ERS1/2 era, starting in 1993, through today (CryoSat-2 and SARAL-AltiKa). The ICESat-2 satellite will continue the sea ice altimeter record and its new technologies will help refine estimates of thickness and potentially also of the snow depth on sea ice (Lawrence et al 2018), particularly if the radar altimeter on CryoSat-2 continues to operate beyond the start of the operative phase of the laser altimeter on ICESat-2.

Information on snow thickness over Arctic sea ice is crucial, since snow strongly affects thermodynamic ice growth, melt pond formation, and radiative transfer. Information on snow thickness on Arctic sea ice is still sparse, although the amount of data has increased recently due to new in situ, airborne and satellite surveys (e.g. Laxon et al 2013, Renner et al 2014, Haas et al 2017, Rösel et al 2018). Reanalysis data can fill in where observations are lacking, but intercomparisons with in situ measurements have shown that they do not always match well (Boisvert et al 2018). Autonomous buoys with snow thickness sensors (ultrasonic sensors or active thermistor chains as used in IMBs) give quantitative time series data with a higher temporal resolution, and usually with a smaller footprint than airborne and satellite surveys (e.g. Brucker and Markus 2013, Maaß et al 2013, 2015). These instruments can directly measure changes in ice thickness and delineate surface and bottom melt (Perovich and Richter-Menge 2015). However, to collect representative data for a region requires many buoys. So far, the total amount of buoys at a given time in the Arctic is still rather limited, and the distribution often not very regular, so that vast areas might be without buoys at a given time (figure 2). Initiatives such as the International Arctic Buoy Programme (Rigor et al 2016, 2000) help coordinate deployments of sea ice buoys. Sophisticated buoy setups that combine measurements related to surface energy and mass balance have been developed recently (e.g. Jackson et al 2013, Wang et al 2014, 2016, Hoppmann et al 2015), as well as buoys that float and have the potential to 'survive' a regional ice-free summer (see e.g. Polashenski et al 2011). Such new developments will contribute to denser and accurate autonomously collected sea ice datasets. Understandably, relatively cheap drifter buoys with GPS but not much other sensors are the most frequent buoys used in the Arctic, while more sophisticated and multi-sensor setups are relatively rare. However, for collecting information on sea ice thermodynamics more ice mass balance buoys (IMBs) would be necessary, along with global data archives where all respective data would be accessible for the scientific community.

Zoom In Zoom Out Reset image size

Recent Arctic expeditions have made increased use of new technology with mobile sensor platforms. These include autonomous underwater vehicle, unmanned airborne vehicle, remotely operated vehicle and instrument sledges. Some of these systems are still at the development stage, but with new systems becoming standard tools, it will be possible to obtain high quality sea ice data over larger spatial scales in the near future. There are still fewer in situ sea ice surveys undertaken in winter months than summer months, but recent (Granskog et al 2016, 2018, Assmy et al 2017, Rösel et al 2018) and planned interdisciplinary expeditions (e.g. the Multidisciplinary drifting Observatory for the Study of Arctic Climate (MOSAiC), www.mosaicobservatory.org) are adding to the existing winter data and providing insights into the changing Arctic sea ice system.

Ice age, also often used as a climate indicator (e.g. Perovich et al 2018, Stroeve and Notz 2018) is now less correlated with ice thickness than it used to be (Tschudi et al 2016a), since the older ice classes have become thinner. First-year ice with little snow cover can reach a thickness similar to second-year ice (Notz 2009), which has also been found in a recent field experiment in the Arctic Ocean (Granskog et al 2016). This can be a challenge for deriving ice classes from ice thickness data (Hansen et al 2014). Distinguishing first- and second-year ice by means of its thickness was probably also not always trivial previously, and is complicated using satellite microwave radiometry as well (Comiso 2006), but in total the range of thicknesses has become narrower, with the thickest ice classes becoming rare, making this more difficult. Assessing the age of ice can be supported by information on its physical properties, for example via their signature in satellite microwave scatterometry (Belmonte Rivas et al 2018), or by tracking ice movement over time (Tschudi et al 2016a). The Lagrangian tracking of ice parcels used here is subject to errors in the motion tracking, especially for regions with high spatial variability in ice dynamics, types and age within the tracked ice parcels which triggered attempts for a revision of the sea-ice age retrieval method (Korosov et al 2018). But the general good agreement between the ice age fields and ice thickness and multi-year ice extent estimates (Maslanik et al 2011) make this pan-Arctic dataset still a valuable tool to assess sea ice age on large spatial scales.

The Lagrangian ice age fields are derived from sea ice motion products (e.g. Lavergne et al 2010, Tschudi et al 2016b), which use feature tracking algorithms to estimate sea ice displacement in satellite imagery. The most commonly used source is passive microwave because of its all-sky capability. However, the low spatial resolution is a significant limitation on the precision and accuracy of the estimates and important small-scale features such as lead formation and ridging cannot be detected. The growing catalogue of SAR sensors has the potential to routinely provide daily, near-Arctic wide coverage at much higher spatial resolutions (e.g. RADARSAT Constellation), though the passive microwave record will still be valuable for long-term context.

Scientists are monitoring and studying sea ice as it drifts through Fram Strait, the area between Greenland and Svalbard, a major gate and only deepwater connection connecting the Arctic Ocean to the other world oceans (Kwok et al 2004, Spreen et al 2009, Smedsrud et al 2011, Hansen et al 2013, Renner et al 2014, Smedsrud et al 2017). Through this gate, most of the sea ice that is exported from the Arctic Ocean passes through here. Findings from here can give information on a larger area 'upstream' (Kwok 2009, Hansen et al 2013, Krumpen et al 2016). With current technology and infrastructure, continuous monitoring at such gates is only possible with moored instruments (e.g. upward looking sonars for measuring sea ice draft), satellite remote sensing, and local meteorological information. Crucial gaps in this context are or can be that there are only few or no observation sites at a gate, and potential data gaps due to instrument failures. Additional limitations arise from current remote sensing capabilities for studying the complex dynamics at such gates and a mis-match in the spatial resolution of the applied data sources (e.g. Ricker et al 2018). Given the large across-strait variability in sea-ice type, thickness and motion, SAR-based estimates of the sea-ice motion are possibly the best way to proceed (Smedsrud et al 2017) provided enough in situ data, e.g. from buoys, allow continued evaluation.

Information about ice thermodynamics with high temporal resolution is often retrieved from IMBs and in situ measurements. Uncertainties are related to the question how homogenous the ice cover is, or in other words how representative a IMB site is for the region it is placed in. Depending what technology in IMBs is used, detection of snow ice and superimposed ice might be not trivial, and this can lead to an overestimation of the snow thickness and an underestimation of the ice thickness.

Drifter buoys with iridium transmitted GPS positions have a better accuracy than Argos-based buoys without GPS earlier, but ice motion can be underestimated due to motion with changing drift direction between individual position recordings. For avoiding that, we recommend position recordings at least once an hour. Sophisticated methods have been developed that use iridium transmission GPS beacons to examine sea ice motion following methodology developed for atmospheric dynamics (e.g. Lukovich et al 2015). Methods of floe tracking from satellite products are well suited to local and regional scale motion tracking (e.g. Komarov and Barber 2014, Itkin et al 2018) and detection and modelling of sea ice hazards to marine navigation (e.g. Barber et al 2014). A key gap is to develop an improved understanding of sea ice dynamic growth as a function of sea ice ridging and rubbling. Essential for this is also improved information on the spatial distribution and variability of surface roughness. The process of dynamic growth is expected to result locally and regionally in thicker ice classes even though thermodynamic growth will be reduced (see case study in Itkin et al 2018).

Knowledge and data gaps continue to limit the certainty of predictions concerning how the ice-associated ecosystem and its coupling to the pelagic and benthic food webs will respond under a changing climate. The relative ease of accessing coastal landfast ice versus the central Arctic ice-pack has resulted in biogeochemical observations being focused in key locations around the periphery of the Arctic Ocean, leaving large data gaps for vast areas of the shelves and central basins (e.g. Leu et al 2015). Similarly, logistical considerations, as well as a focus on the biologically productive period, have skewed the seasonal observational base towards spring-summer when environmental conditions are more favorable to field studies. The few autumn–winter studies that do exist (e.g. Niemi et al 2011, Niemi and Michel 2015, Assmy et al 2017, Melnikov et al 2016, Olsen et al 2017) reveal interesting ecological features of the ice-associated ecosystem, calling for a better characterization of biogeochemical processes during this time of year. Little remains known about the distribution and bloom development of sub- and under-ice primary producers and there are few within-ice primary production estimates.

Baseline information on multi-year sea ice communities encompassing biomass, productivity and biodiversity, and role in the cycling of carbon and other elements, is also an important knowledge gap which requires urgent attention given the rapid decline in Arctic multiyear ice. With multiyear ice potentially being more productive than traditionally assumed (Lange et al 2017), there is a critical need for rate measurements in multiyear ice. The likelihood of changing ice and ocean conditions enhancing the potential for toxin-producing algal blooms in the Arctic highlights a need for a better understanding of the presence, distribution, and recurrence of these species and their influence on Arctic food webs, as well as a need to improve the reporting of toxin occurrences across the Arctic. Repeat observations of higher trophic levels are also required, including fishes and marine mammals, in order to constrain their ecological responses and distributional changes in relation to the changing sea-ice cover.

Owing to the interacting effects of light and nutrients as the main factors controlling primary production at the sea ice interface and under the ice, uncertainties associated with observations and projections of the physical variables by which they are controlled, confound the capacity to predict the future structure and function of the sea-ice ecosystem. On Arctic shelves, sea ice and sea ice export contribute significantly to structure spatially diverse productivity regimes, also influenced by nutrient inventories and dynamics (Michel et al 2015, Tremblay et al 2015). As these different productivity regimes respond differently to on-going and future Arctic change (Ardyna et al 2011), a key challenge remains in resolving meso- to large-scale heterogeneity across the shelves and basins, for primary production estimates. In addition, poorly constrained estimates of the contribution of ice algae and phytoplankton to primary production and transfers to the food web, including harvest resources, require further attention. Furthermore, these uncertainties increase when predicting the responses at higher trophic levels, due to the added complexity of interactions at the species, population, community and ecosystem level.

Beyond gaps connected with sea ice ecosystem observations, it is important to mention that there is also a rather limited amount of modelling studies that address both the complexity of the physical system, and the wide range of components of the ecosystem (e.g. Duarte et al 2015, 2017).