In this chapter, the committee identifies a number of key challenges that currently limit our ability to understand and predict sea ice evolution. These key challenges were motivated by several key science questions that the committee considers crucial to answer in order to improve our understanding and ability to predict sea ice on seasonal to decadal scales (Box 2.1). The key science questions are intended to serve as a distillation of the more detailed discussions that follow.
Many of the overarching challenges are not necessarily unique to the topic of sea ice prediction. However, their acknowledgment is important because they need to be considered in formulating strategies to advance predictive capabilities.
Key Challenge: Treating Sea Ice as Part of the Global System
A key to advancing our understanding and predictive capabilities is the treatment of the sea ice cover as an integral part of the complex Arctic system, which in turn is an integral element of the global system. Adding to this complexity is the broad range of human activities that not only influence the Arctic system, but are also influenced by it.
The Arctic sea ice cover is part of the larger Arctic system, which in turn is part of the larger global system, with interactions across all components. Arctic sea ice is a key element of these systems, influencing and influenced by change across a wide range of temporal and spatial scales (Figure 2.1). A myriad of feedback mechanisms link the atmosphere, sea ice, ocean, seafloor, and land, many of which are not yet fully understood (Maslowski et. al., 2012). For example, winds and ocean currents can alter the distribution of sea ice. These changes in the sea ice cover can then affect large-scale circulation patterns in the atmosphere (e.g., Deser et al., 2010; Francis and Vavrus, 2012) and ocean (e.g., Guemas and Salas-Melia, 2008), which in turn may impact weather, fisheries, and the global climate system. Other system components affecting and affected by sea ice cover include biological and chemical processes (e.g., Simpson et al., 2007; Kelly et al., 2010b; Deming and Fortier, 2011; ). In addition, bathymetry plays a role in processes such as sea ice formation and evolution (Jakobsson et al., 2012; Nghiem et al., 2012).
KEY SCIENCE QUESTIONS
• What are the implications of the recent dramatic shifts in the Arctic from predominantly multiyear ice to first-year ice, and how will the associated complexities of this regime shift affect sea ice variability and predictability?
• In a rapidly changing Arctic regime, how will forcings and couplings between the various components of the ocean, atmosphere, cryosphere, and seafloor and land systems modify or influence the processes governing the characteristics of the sea ice cover?
• What are the impacts of extreme events and feedback mechanisms on Arctic sea ice evolution and our ability to predict it?
• How will the changing Arctic sea ice characteristics and dynamics affect stakeholders on a variety of timescales, including prediction requirements?
Adding to the complexity of the natural system is a range of human interactions. Humans not only influence the Arctic (e.g., impacts of oil spills, ship discharges, and land and sea greenhouse gas emissions), they are also influenced by it (e.g., decisions to pursue commercial shipping routes and offshore developments in response to a changing ice cover, long-term strategies for Arctic security, and development of future Arctic monitoring systems). The Arctic sea ice cover, therefore, cannot be viewed in isolation, as it is constantly responding to a host of regional and global forces, and these are in turn directly influencing the ice cover’s seasonal and decadal presence or absence. Treating the Arctic as an integrated whole and a vital component of the global system is necessary to significantly advance our understanding of the sea ice cover as part of a complex system.
Key Challenge: Impacts of the Regime Shift of Arctic Sea Ice
Understanding how the recent regime shift in the Arctic sea ice cover, resulting in a significant reduction in the amount of multiyear ice compared with first-year ice, affects the processes governing the atmosphere-sea ice-ocean system and the models and observations used to study and predict Arctic sea ice dynamics.
The recent decline in the extent of Arctic summer sea ice has resulted in a dramatic shift in its composition (Figure 2.2). First-year ice is becoming more prevalent within the Arctic Basin, reducing the size of the multiyear ice pack (e.g., Maslanik et al., 2011). Furthermore, the multiyear ice that does remain is younger and thinner (Haas et al., 2008; Comiso, 2012). This rapid change to a new state is likely to have important implications for sea ice variability (Goosse et al., 2009; Hutchings and Rigor, 2012) and, ultimately, predictability (Holland et al., 2011). For instance, recent research suggests that first-year ice is not only more
FIGURE 2.1 The Arctic system is made up of various components including a complex network of process interactions, interdependent feedbacks, and thresholds. There are many interconnections among system components and important changes in one component may influence numerous other parts of the system, including sea ice. SOURCE: Adapted from the Study of Environmental Arctic Change.
susceptible to summer melt (Perovich and Polashenski, 2012) but it is also likely to be more easily ridged, ruptured, and transported by winds (Rampal et al., 2009).The delay in the formation of shore-fast ice as well as its reduced stability and overall duration are also significant changes, with implications for stakeholders along the Arctic coast (Mahoney et al., 2007).
Some of the greatest changes to the Arctic Ocean are occurring in the Chukchi and Beaufort seas, where the increased summer ice retreat has created a substantially enlarged marginal ice zone (Holland and Stroeve, 2011). As a seasonal feature contained within the Arctic Basin,
this region of transition between the ice-covered and open ocean is expected to have different governing processes than the marginal ice zones in the Greenland Sea or Southern Ocean. There are also indications that the extent of the North Atlantic marginal ice zone has increased over the past several decades (Strong, 2012).
It is important to understand how the transition from a multiyear to first-year ice regime affects (a) the realism of current numerical models and their parameterizations, (b) the power of statistical prediction methods, (c) the processes governing the atmosphere-sea ice-ocean system, and (d) instrument design and observational strategies. Currently, sea ice models’ treatment of ice dynamics and thermodynamics employs parameterizations that were often developed using insights gained from observations taken in a primarily multiyear ice regime. For example, many sea ice models use ice dynamic and ice thickness distribution treatments based on the Arctic Ice Dynamics Joint Experiment that took place during the 1970s (e.g., Thorndike et al., 1975; Hibler, 1979). Model parameterizations based on the Arctic system’s behavior in the past may not apply in the new state. Moreover, it is likely that if, as expected, the substantial ice retreat continues and the remaining ice transforms to a largely seasonal character, the oceanic and atmospheric circulation and thermodynamic structure will respond to the changes in the surface state, affecting large-scale patterns. The regime shift may also cause changes in physical and biochemical processes that are not or have not been adequately accounted for in current models. These complex interactions need to be simulated with sufficient accuracy to robustly project changes in the system.
A number of smaller-scale processes also need to be parameterized differently in first-year and multiyear ice because of inherent variations in their morphologies. For instance, recent work by Perovich and Polashenski (2012) conducted near Barrow, Alaska, found that once surface melt begins, seasonal ice albedos are consistently less than those for multiyear ice. This finding suggests that the shift from a multiyear to seasonal ice cover has significant implications for the heat and mass budget of the ice through a strong positive feedback on melting, and for primary productivity through an increase in the amount of sunlight penetrating into the upper ocean (Arrigo et al., 2012).
A shift from predominantly multiyear to mostly first-year ice will also affect the skill of statistical prediction methods. These methods use statistical relationships determined from past system behavior to predict the future state of the ice cover. An underlying assumption is that these relationships are stationary. However, given the transition to a thinner and increasing first-year ice pack, statistical relationships that have provided predictive skill for the past may no longer be valid.
The fundamental shift in the ice regime is also likely to have significant implications regarding the design of observation studies and networks, including the instrumentation. As is the case with sea ice models, most Arctic observational system designs and instruments were “developed” in the era of a predominantly multiyear sea
FIGURE 2.2 Sea ice is shifting from a composition of predominantly multiyear ice to include increasing amounts of first-year ice. Sea ice age distribution in March 1988 (upper left) compared with 2012 (lower right) illustrates the extensive loss in recent years of older ice types. Ice age is determined using satellite observations to track ice parcels. Older multiyear ice tends to be thicker and thus more resistant to forcing than younger first-year ice. SOURCE: James Maslanik, Charles Fowler, and Mark Tschudi.
ice cover and thus may need retooling to address thinner, more mobile ice conditions.
For example, many buoys deployed in coordination with the International Arctic Buoy Programme were largely designed for placement on multiyear sea ice floes to maximize their longevity. There are recent examples of work dedicated to transitioning the design of ice-based instruments for deployment in the seasonal ice zone, where they tend to be more vulnerable (e.g., Polashenski et al., 2011). New sustained
measurement strategies can also take advantage of innovative technologies such as unmanned aerial systems and autonomous underwater vehicles to answer specific questions without the risks of deployment on seasonal sea ice or in open water.
Key Challenge: Identifying Diverse and Emerging Stakeholder Requirements
As the Arctic is being transformed by globalization and climate change, new stakeholders with additional and more sophisticated requirements are emerging. Clearly defining these diverse needs as they relate to seasonal to decadal sea ice prediction is crucial to inform the future directions of modeling, observations, and overall research.
The term stakeholder covers a wide range of communities with interests in the development and application of Arctic sea ice prediction across seasonal to decadal timescales. The term includes indigenous residents on the Arctic coast, scientific researchers, commercial users (e.g., fishing, shipping, natural resource development, and marine tourism.), and naval and coast guard planners. Not surprisingly, this diverse group of stakeholders has a broad and evolving set of needs and, hence, sea ice prediction requirements. In addition to ice type (e.g., multiyear/perennial vs. first-year/seasonal ice), extent, thickness, and motion, there is growing interest in the ability to predict snow cover depth, melt onset, freeze-up/ice season length, and the size and shape of ice floes (e.g., Ray et al., 2010). For example, earlier snowmelt onset date and thinner snow cover during winter are important parameters for an early snowmelt, which may prematurely destroy subnivean lairs subjecting ringed seal pups to adverse weather and increased predation (Kelly et al., 2009; Hezel et al., 2012; see Figure 2.3).
Understanding stakeholders’ needs for seasonal to decadal sea ice prediction is crucial to guiding the future directions of modeling, observations, and overall research. Effectively quantifying and communicating the inherent limitations in sea ice predictability is also needed to establish reasonable stakeholder expectations. Significant issues that will need to be overcome include: (1) how to aggregate stakeholder input; (2) how to prioritize among stakeholder interests and needs; (3) how to communicate the limitations of modeling and forecasting to users and policy makers; (4) how to maximize the utility of sea ice predictions containing uncertainties; and (5) how to assess opportunities for overlapping interests and efficiency of efforts in meeting stakeholder needs.
A realistic expectation for the future state of Arctic sea ice is needed to manage risks, deploy limited resources, and plan long-term infrastructure. Plausible future sea ice scenarios projected by fully coupled, large-scale global climate models (GCMS) can be useful for strategic planning. For instance, they are used by federal and private entities and commercial shipping companies looking to understand future navigation seasons on potential Arctic shipping routes. GCMs are also used by federal agencies to study the plausible futures of endangered
FIGURE 2.3 Where sufficient snow accumulates on the sea ice cover, ringed seals may excavate snow caves (lairs) above the breathing holes to protect the pup from adverse weather and predation. Early snowmelt onset causes premature lair abandonment exposing pups to hyperthermia and predation. Images courtesy of Brendan Kelly (left) and Juha Taskinen (right).
species and the possible merits of regulatory strategies (e.g., NOAA, 2010). However, better characterizing the uncertainties through continued intercomparisons of model outputs as well as validation with sea ice observations is necessary to enhance confidence in using this information for decision making.
Although projections from GCM simulations fulfill some requirements for planning, regional forecasts and seasonal predictions at shorter timescales and finer spatial resolutions are more useful to offshore drilling operations, marine transportation systems, and subsistence hunters. Many of these requirements call for timely predictions that resolve the highly varying ice characteristics that occur near the coastline, demanding accuracy of the prediction within a few hundred kilometers (or tens of kilometers) of shore and within the marginal ice zone. Modelers are developing methods to address these needs—by either nested grids or regional high-resolution models—but they face challenges related to the extreme computational demands, lack of suitable initialization conditions, inadequate projections of atmospheric and oceanic forcing, and insufficient software and resources to manipulate and store the enormous datasets produced. Parameterizations that have been developed for relatively coarse resolutions, such as the viscous-plastic sea ice rheology (Hibler, 1979), may not be applicable at very fine spatial scales. This may necessitate additional model developments such as the incorporation of more detailed representations of sea ice dynamic components (e.g. Hopkins and Thorndike, 2006; Sulsky et al., 2007) into large-scale climate models.
Key Challenge: Competing Approaches to Seasonal Sea Ice Prediction
Although limitations in the various approaches used to generate seasonal forecasts are generally acknowledged, there is a lack of quantitative information about their relative strengths and weaknesses.
Three approaches are used to provide seasonal ice forecasts, including (1) statistical algorithms (e.g., Drobot et al., 2006); (2) coupled ice-ocean models driven by prescribed atmospheric forcing (e.g., Zhang et al., 2008b); and (3) more recently, coupled atmosphere-ocean-ice models. All three approaches have contributed to the SEARCH (Study of Environmental Arctic Change1) Sea Ice Outlook.2 Each of the methods has acknowledged limitations, but the relative accuracies, strengths, and weaknesses are poorly known. For example, statistical relationships based on past observations may not be valid in a future Arctic dominated by thinner ice (Holland and Stroeve, 2011).
In coupled ice-ocean prediction systems, atmospheric forcing is usually based on an ensemble of past years’ winds and temperature fields derived from atmospheric reanalyses. However, the statistics of past forcing fields may not be applicable to the new Arctic state, because the preponderance of thin ice, large-scale changes in surface characteristics, and resulting shifts in atmospheric patterns all affect surface-atmosphere exchanges. Additionally, important feedbacks between the atmosphere and ice that influence the sea ice evolution are not accounted for by using historical atmospheric conditions. In coupled atmosphere-ocean-ice model predictions, atmospheric forcing is free to evolve consistent with the underlying distributions of sea ice and ocean temperatures, but the simulations of sea ice evolution suffer from systematic biases. This causes the models to drift to their preferred (and biased) climate state once constraints by observations are removed. Methods to bias correct (or “de-drift”) forecasts with coupled models have been explored (e.g., Yeager et al., 2012, for ocean conditions). Alternatively, “anomaly-initialization” in which observed anomalies are added to the model'’s preferred climate state has been used in a number of decadal prediction studies (e.g., Pohlmann et al., 2009).
The relative benefits of these various approaches have not been assessed for sea ice prediction. Determinations of their prediction accuracy are challenged further by observation-based uncertainties in the sea ice variables that are to be predicted. This is especially the case when the predictions extend to new variables (e.g., sea ice thickness, morphology, and stability) that are of particular and growing interest to stakeholders.
Key Challenge: Observational Requirements for Seasonal Sea Ice Prediction
Seasonal sea ice prediction capability depends on adequate knowledge of initial ice-ocean conditions, even though the specific requirements associated with “adequate knowledge” have yet to be established. This challenge is compounded by the need for a fast turnaround in acquiring and accessing observational data.
There has been a dearth of experiments performed to systematically evaluate the sensitivities of sea ice predictions to the type, quality, density, and frequency of observations of the Arctic atmosphere, ocean, and sea ice. These sensitivities arise from the specification of the initial state of the sea ice and its drivers in numerical models.
During the forecast period, atmospheric variations, which are largely unpredictable beyond a week or two, are expected to have a significant influence on sea ice and will limit its seasonal predictability. Instead, sea ice predictability on these timescales resides in the initial ice and ocean state. Factors believed to be among the most important for predicting sea ice behavior on the seasonal scale include accurate knowledge of (1) sea ice conditions at the start of the season (particularly the ice thickness distribution and the partitioning between seasonal and multiyear ice) and (2) upper ocean conditions, such as ocean mixed-layer heat content.. However, the relationship between prediction skill and the uncertainty in each of these factors is poorly known. The importance of initial values of other variables—such as snow on sea ice, ocean temperature and salinity profiles below the ice, and ocean current distributions—is poorly understood, but may be considerable. The accuracy requirements for bathymetry, which constrains ocean currents and controls the distribution of warm and cold water masses, are largely unknown (Jakobsson et al., 2012).
There is little information on what observational quality, spatial density, location, and accuracy are required for different variables to realize useful predictive power. Compounding the problem is the fact that observations of the atmosphere, sea ice, ocean, and seafloor are made from vastly different sensors, including in situ, airborne, and satellite instruments, along with a variety of methods. Each approach brings with it a specific set of characteristics, which often go undocumented.
There has been little if any effort to establish and broadly apply measurement protocols and standard definitions. For instance, even for one of the most fundamental parameters such as air temperature, there is no existing national or international protocol for predeployment calibration, maintenance of measurement stability during deployment, post-deployment calibration, and cross calibration among different sensors and different algorithms (e.g., different algorithms to measure ice surface temperature from different satellite sensors).
Adequately acquiring and making observations available within several days of the beginning date of the model simulation are necessary if the observations are to be useful for initializing operational seasonal model predictions. Ice extent is available within a day of satellite data acquisitions from multiple satellite sensors (passive microwave, active microwave, and
multispectral sensors). Recent efforts undertaken by the National Aeronautics and Space Administration (NASA) IceBridge project (Kurtz et al., 2012) and the Sea Ice Outlook,3 for example, have demonstrated the ability to deliver observations of sea ice thickness and snow depth within a period of weeks from airborne platforms. This achievement was the result of a coordinated effort backed by dedicated resources (i.e., measurement platform, instruments, personnel, and funding resources).
However, near-real-time access remains problematic for many other types of data, such as ocean temperatures, salinities, and ice draft collected from seafloor moorings. Moreover, ocean currents, the distribution of warm and cold waters, and ocean mixing are controlled, guided, or restricted by seafloor characteristics. About 89 percent of the Arctic Ocean still needs to be surveyed to determine where interpolation or extrapolation may result in unrealistic bathymetric features (e.g., false seamounts) or where important structures (e.g., canyons) have been missed (Jakobsson et al., 2012). In general, prioritizing data acquisition with sufficient temporal and spatial coverage and timely accessibility to observations is critical if they are to play an effective role in seasonal, operational sea ice prediction.
Coupled climate models calculate their own atmospheric and oceanic variables from the basic laws of physics, providing the thermodynamic and dynamic drivers of sea ice evolution. Considering the number of variables and interactions that affect the ice cover, it is a tremendous achievement that the models simulate the observed ice behavior as well as they do. Any errors in the variables, processes, or feedbacks involved in these calculations, however, will be propagated and integrated by the ice over time and result in an unrealistic representation. A key challenge is simulating realistic atmospheric and oceanic conditions, which in turn depend on assumptions about future trends in carbon dioxide emissions, aerosol loading, changing surface characteristics, etc. Related to this challenge is the difficulty in determining which processes in a particular model are responsible for unrealistic aspects of sea ice simulations, especially systematic biases such as those responsible for the large model spread in Figure 1.3.
A key challenge in coupled climate models is the capability to realistically simulate atmospheric and oceanic conditions of relevance to sea ice variability, including the identification of model processes that contribute to unrealistic forcings and feedbacks.
Phenomena such as El Niño-Southern Oscillation (ENSO), the Arctic Oscillation (AO), the Pacific Decadal Oscillation, and the Arctic Dipole are known to affect the ice cover through their influence on ice transport, storm tracks, and heat transport (e.g., Rigor et al., 2002; Wang et al., 2009; Ogi et al., 2010). What drives these
atmospheric patterns into their positive and negative phases in any given year or sequence of years, however, is not well understood. Although models generally simulate these major modes of variability, studies reveal potentially important discrepancies in the statistics of their variations (e.g., Stenchikov et al., 2006). Recent studies have suggested that phasing of the AO may be related to stratospheric influences in some situations (Black, 2002; Cohen and Jones, 2011) and to surface forcing in others (Overland and Wang, 2010). Thus models need to include a realistic representation of the stratosphere and its interaction with the tropospheric circulation.
The present capability to predict, in detail, the large-scale modes of atmospheric variability is limited, and in some cases there may be little deterministic predictability beyond a few weeks. Experiments addressing such changes in large-scale atmospheric modes are challenged by natural variability, because many ensemble members can be required to detect significant changes in the circulation (Bhatt et al., 2008; Deser et al., 2012). Despite these challenges, models need to realistically simulate the statistics of these atmospheric variations, including their spatial patterns, frequency of occurrence, and response to varying forcings, if they are to accurately simulate sea ice variability on decadal scales. Whether fluctuations in spatial extent, intensity, and frequency of these large-scale oscillations will change as greenhouse gases continue to accumulate is a key open question. Evidence does suggest that high-latitude surface changes—which include changes in sea ice cover that may affect wind patterns (e.g., Liu et al., 2012), water vapor content, and cloud amount (e.g., Winton, 2006; Kay et al., 2012)—can then feed back onto the ice cover (Overland et al., 2012;Wu et al., 2012) and steer ocean currents. Capturing these feedbacks in coupled models is critical if decadal predictions are to be successful.
As evidenced in summer 2007, extreme events (e.g., anomalous winds as one of several key factors) may combine with preconditioning and ice-albedo feedback to result in abrupt change (e.g., a large decrease of sea ice in a short time) (Haas et al., 2008; Perovich et al., 2008; Zhang et al., 2008a; Lindsay et al., 2009; Ogi and Wallace, 2012) that can have decadal impacts. For example, drastic loss of perennial sea ice owing to persistent wind patterns in 2005 and 2007 (AMAP, 2011) may influence the long-term sea ice trends. Models do simulate extreme events of this type (e.g., Holland et al., 2006) but the realism of how simulated extreme events modify key parameters needs to be further assessed.
Another characteristic that highlights the interconnectedness of the Arctic system is the influence of the Atlantic and Pacific oceans on the Arctic Ocean. Relatively warm water masses from the Atlantic and Pacific enter the Arctic Ocean, and because they are saltier than the surface waters, reside below the mixed layer. The Arctic Ocean’s present stratification, resulting primarily from the vertical salinity profile, largely limits heat transfer to the ice cover from the deeper layers. These deeper layers contain vast quantities of heat that could melt all of the sea ice relatively quickly (e.g., Alexeev et al., 2011). As the Arctic transitions to a state
dominated by first-year ice and as atmospheric circulation patterns potentially change in response to sea ice loss and other factors, the stratification of the upper-ocean layers could also change, modifying the amount of Atlantic and Pacific layer heat that affects the ice cover. Moreover, additional solar energy absorbed in areas where ice has retreated will also affect mixed-layer characteristics (Perovich et al., 2007). Owing in part to insufficient vertical resolution in modeled oceans, processes related to the mixed layer and stratification are not well captured by models. Additionally, lateral oceanic transports affecting, for example, the net heat flux from the North Atlantic into the Arctic Basin, are also often poorly simulated in current climate models.