Solar activity affecting Earth and its planetary neighbors encompasses a wide range of phenomena, from the steady solar wind and the interplanetary magnetic field to extreme events like solar flares, coronal mass ejections (CMEs), and solar energetic particle (SEP) events. These space weather phenomena interact in complex ways with planetary magnetospheres and atmospheres. On Earth, we see the results in the dancing lights of stunning auroras and in less frequent but sometimes severe disruptions to telecommunications, navigation, and energy infrastructure.

Forecasting conditions throughout the heliosphere (the region influenced by the solar wind), understanding the variety of Sun-Earth interactions, and predicting arrivals of space weather events—both benign and potentially hazardous—are a grand challenge.

Solar flares emit electromagnetic radiation that spreads in all directions. In contrast, the propagation of CMEs and SEP events depends on their source location on the Sun and on the heliospheric magnetic field, which is carried outward by the solar wind. The impacts these events have on magnetosphere systems further vary depending on particle energies and intensities in SEPs and on particle speeds and the magnetic field orientation in CMEs. The Sun-Earth challenge thus requires tracking and predicting conditions—from routine and quiet to rare and extreme—across tens of millions of kilometers of interplanetary space.

This tracking and prediction is powered by petabyte-scale datasets from solar observatories and spacecraft measurements that provide rich observational archives. Researchers use these data to deduce physically meaningful quantities describing the heliosphere and to identify patterns to distinguish quiet from active conditions. The resulting insights not only answer fundamental science questions but also provide critical prediction time frames needed by space weather forecasters.

Even with all these data, the enormity of space between the Sun and Earth presents a major obstacle to our predictive capabilities. Another obstacle is that the data are obtained by different instruments operating at different locations and times. These factors combine to create a unique data sparsity challenge that complicates large-scale analysis.

These fundamental issues—the massive yet still insufficient supply of data available, the extreme differences in the scales of the processes we must illuminate, and the need for actionable predictions—suggest opportunities for artificial intelligence (AI) and machine learning (ML) to complement traditional physics-based analytical approaches [Camporeale, 2019]. In a series of workshops—insights from which inform the discussion below—scientists explored such opportunities and how they can advance heliophysics research and operational space weather forecasting.

The Need for Space Weather Forecasting

Space weather events can have significant impacts on infrastructure and humans. They can disrupt satellite operations (e.g., by enhancing atmospheric drag on satellites), damage electronics in space, interfere with radio communications and GPS, and even affect power grids (e.g., through geomagnetically induced currents) during the most severe events. They can also pose risks to people, especially astronauts beyond the protection of Earth’s atmosphere and airline crews and passengers on long-distance polar flights, during which exposure to energetic particles is elevated. Forecasting offers a first line of defense in preparing for or preventing damaging and hazardous effects of space weather.

In assessing major CMEs, forecasters consider whether and when events will reach Earth and whether they will trigger geomagnetic storms and substorms. For SEP events, predictions must include arrival times, peak intensities, durations, and energy characteristics.

Predicting extreme space weather phenomena is vital, but equally important is forecasting periods when no significant activity is expected, which is critical information for satellite operators and other stakeholders. Making such predictions requires understanding physics spanning 8 orders of magnitude in space and time, from subsecond processes in Earth’s magnetic environment to multiday solar eruptions propagating across the 150 million kilometers between the Sun and Earth (Figure 1) and long-term interactions at scales associated with the 11-year solar cycle.

In addition to operational forecasting, these challenges are fundamental in heliophysics research. Such research includes work to reveal how the Sun generates its magnetic field, how solar wind accelerates and evolves, how planetary magnetospheres respond to external forcing, how particles are accelerated, and how energy transfers across multiple scales and regimes.

Unique Challenges in Heliophysics

Modern AI and ML algorithms excel at analyzing well-curated, extensive datasets that include millions of training examples. For example, AI-aided terrestrial weather forecasting relying on continuous, high-resolution coverage from thousands of ground stations, weather balloons, and satellites has advanced dramatically in recent years.

Heliophysics, however, presents a unique and somewhat opposite scenario. Fewer than a dozen spacecraft monitor Earth’s magnetosphere, a region spanning tens of Earth radii (about 6,371 kilometers). Solar wind observations are even sparser, with just a handful of monitors scattered across the space between the Sun and Earth. This fundamental scarcity poses a challenge for data-driven approaches, which typically depend on abundant observations that are well distributed in space and time to produce trustworthy (i.e., generalizable and reproducible) models.

Data sparsity is further compounded by the relative rarity of intense space weather phenomena such as CMEs, major geomagnetic storms, and extreme substorms, which occur only a few times per solar cycle. Most heliophysical observations capture quiet, low-activity conditions when the solar wind is steady and magnetospheres are calm. Standard ML approaches trained on such imbalanced datasets may achieve high statistical accuracy by simply predicting a “nothing-will-happen” outcome but completely fail when extreme events occur.

Although solar eruptions and geomagnetic storms are relatively rare, they exhibit recurring patterns and consistency in their physical drivers. This regularity suggests that historical observations, when properly clustered and analyzed, can be used to enhance prediction capabilities. The challenge therefore lies in extracting meaningful patterns from sparse measurements of rare events while avoiding models that work well for average conditions but fail when they matter most [Chu et al., 2025].

AI Solutions for Data Sparsity

Heliophysics research employs clever approaches to extract maximum information from the limited available observations. One strategy is to mine multidecade observational records from various satellites and to match and group together measurements collected at times with similar solar wind and geomagnetic activity conditions.

This process clusters tens of thousands of data points from similar magnetospheric states. Such clustering enables reconstruction of dynamic features like nightside magnetic field changes during substorms [Stephens et al., 2019] and the presence of near-Earth magnetotail reconnections [Angelopoulos et al., 2020].

Another, more universal approach is to embed fundamental physical laws directly into ML models through physics-informed neural networks [Raissi et al., 2019], ensuring that predictions respect physical reality even when training data are limited. Data assimilation techniques used in weather forecasting similarly blend sparse observations with physics-based simulations and update models as new measurements arrive.

These methods converge on a common theme: building gray box models (so named because they’re less opaque than black box models) that are data driven but grounded in physically real constraints. For data-starved applications, hybrid approaches can outperform purely data-driven or purely physics-based methods [Liu et al., 2025].

Satellite instruments are generating increasingly large solar wind datasets. However, the variables obtained (e.g., solar wind speed and pressure) are highly intercorrelated [Borovsky, 2018], making it difficult to identify which ones truly drive magnetospheric responses. New algorithms are helping to distill datasets without losing critical scientific information [e.g., Camporeale, 2025]. Meanwhile, advanced statistical and ML methods can cut through dataset complexity by reducing dimensionality, identifying causal relationships among variables, and providing clues about dominant drivers.

For instance, information theory provides tools to detect dependencies in complex systems, establish causality, and rank variables that most effectively predict space weather outcomes [Wing et al., 2022]. Such techniques can be paired with other “explainable” tools, such as SHAP (SHapley Additive exPlanations) values, a method inspired by game theory, to pinpoint physical variables (e.g., solar wind speed or magnetic orientation) that drive a prediction [Ma et al., 2023].

Distilling datasets and improving model interpretability help make ML more practical and more scientifically trustworthy and its predictions more robust. But fully trusting ML models in operational environments requires rigorous validation and uncertainty quantification. These models must not only make predictions but also indicate their confidence levels for operational decisionmaking.

When a model forecasts a major geomagnetic storm, operators need to know whether that prediction carries 60% or 95% confidence, for example. Ensemble approaches, in which multiple models provide a range of outcomes, help quantify this uncertainty, while using standardized, well-documented datasets enables fair model intercomparisons.

The research community is developing ML-ready benchmark datasets with consistent formatting and clear metadata to establish such validation procedures [e.g., Angryk et al., 2020]. These resources allow researchers to test new algorithms against common baselines, accelerating progress while ensuring that advances are robust and reproducible rather than artifacts of specific data processing choices.

Notably, one domain in heliophysics that is not affected by severe data sparsity is solar imaging. Decades of continuous, high-resolution observations from the Solar Dynamics Observatory (SDO), which delivers 1.5 terabytes of data every day, have created enormous data archives. Because the Sun drives space weather throughout the heliosphere, these datasets offer an ideal opportunity for use in foundation models, large-scale ML systems trained to learn comprehensive internal representations that can then be easily adapted to specific scientific tasks with minimal additional training.

Surya, a foundation model designed to construct a digital representation of the Sun, represents one such effort. It is still in early development and has yet to be validated, but this approach illustrates how data-rich domains can be leveraged with modern AI techniques to create tools that broadly benefit heliophysics research and space weather forecasting.

Advancing Research and Operational Forecasting Together

In addition to the needs for data and model development and validation, applying AI to address the challenges of heliophysics requires sustained, multidisciplinary collaborations. Fostering those collaborations has been the focus of a series of workshops, with the most recent being 2025’s Machine Learning, Data Mining and Data Assimilation in Geospace (LMAG25) meeting at the Johns Hopkins University Applied Physics Laboratory. The workshops have brought together heliophysicists, machine learning experts, data scientists, and specialists from weather forecasting and applied mathematics to exchange knowledge and establish community standards.

The LMAG forums also serve as gathering spaces for scientists to validate models against diverse datasets, compare physics-based and data-driven approaches, develop performance benchmarks, and discuss how to bridge research and operational requirements. Space weather forecasters need models that are accurate and interpretable and that provide not just statistical metrics but also actionable predictions with known limitations and reliability. Of course, researchers also benefit. These conversations allow them to gain insight into operational constraints that shape how modeling approaches become practical in real-world settings.

LMAG and similar initiatives facilitate direct exchanges among adjacent communities, including by making meeting presentations openly available. These efforts are helping translate cutting-edge AI and ML techniques into practical tools that help protect critical infrastructure and human well-being. They are also deepening our understanding of how the Sun shapes space weather throughout the solar system and its effects—both mundane and major—on Earth.

References

Angelopoulos, V., et al. (2020), Near-Earth magnetotail reconnection powers space storms, Nat. Phys., 16(3), 317–321, https://doi.org/10.1038/s41567-019-0749-4.

Angryk, R. A., et al. (2020), Multivariate time series dataset for space weather data analytics, Sci. Data, 7(1), 227, https://doi.org/10.1038/s41597-020-0548-x.

Borovsky, J. E. (2018), The spatial structure of the oncoming solar wind at Earth and the shortcomings of a solar-wind monitor at L1, J. Atmos. Sol. Terr. Phys., 177, 2–11, https://doi.org/10.1016/j.jastp.2017.03.014.

Camporeale, E. (2019), The challenge of machine learning in space weather: Nowcasting and forecasting, Space Weather, 17(8), 1,166–1,207, https://doi.org/10.1029/2018SW002061.

Camporeale, E. (2025), PARIS: Pruning Algorithm via the Representer theorem for Imbalanced Scenarios, arXiv:2512.06950, https://doi.org/10.48550/arXiv.2512.06950.

Chu, X., et al. (2025), Imbalanced Regression Artificial Neural Network Model for Auroral Electrojet Indices (IRANNA): Can we predict strong events?, Space Weather, 23(5), e2024SW004236, https://doi.org/10.1029/2024SW004236.

Georgoulis, M. K., et al. (2026), Prediction of solar energetic events impacting space weather conditions, Adv. Space Res., in press, https://doi.org/10.1016/j.asr.2024.02.030.

Liu, Y., et al. (2025), Data-driven modeling of electrostatic turbulence by physics-informed Fourier neural operator, Mach. Learn. Sci. Technol., 6(4), 045050, https://doi.org/10.1088/2632-2153/ae19cd.

Ma, D., et al. (2023), Opening the black box of the radiation belt machine learning model, Space Weather, 21(4), e2022SW003339, https://doi.org/10.1029/2022SW003339.

Raissi, M., P. Perdikaris, and G. E. Karniadakis (2019), Physics-informed neural networks: A deep learning framework for solving forward and inverse problems involving nonlinear partial differential equations, J. Comput. Phys., 378, 686–707, https://doi.org/10.1016/j.jcp.2018.10.045.

Stephens, G. K., et al. (2019), Global empirical picture of magnetospheric substorms inferred from multimission magnetometer data, J. Geophys. Res. Space Phys., 124(2), 1,085–1,110, https://doi.org/10.1029/2018JA025843.

Wing, S., et al. (2022), Modeling radiation belt electrons with information theory informed neural networks, Space Weather, 20(8), e2022SW003090, https://doi.org/10.1029/2022SW003090.

Author Information

Savvas Raptis (savvas.raptis@jhuapl.edu), Manolis K. Georgoulis, Mikhail Sitnov, Anthony Sciola, and Simon Wing, Johns Hopkins University Applied Physics Laboratory, Laurel, Md.

Citation: Raptis, S., M. K. Georgoulis, M. Sitnov, A. Sciola, and S. Wing (2026), Vast space, sparse data: An AI answer to twin space weather challenges, Eos, 107, https://doi.org/10.1029/2026EO260188. Published on 11 June 2026.

Text © 2026. The authors. CC BY-NC-ND 3.0
Except where otherwise noted, images are subject to copyright. Any reuse without express permission from the copyright owner is prohibited.

NASA has announced the death of one of its Mars imaging probes, MAVEN, after losing contact with the spacecraft at the end of last year.

[Read More]

Source: Earth’s Future

The space industry is surging. In coming years, nearly 10,000 spacecraft are slated to launch into low-Earth orbit for a variety of purposes, such as global surveillance, space tourism, and satellite “megaconstellations” providing internet service.

Rocket engine exhaust, as well as the burnup of inactive satellites and rocket parts reentering Earth’s atmosphere, releases a suite of pollutants. These chemicals have long been considered to pose little threat to our climate, given the historically small size of the space industry. Now, the sector’s rapid growth will send its emissions skyrocketing—but scientists don’t yet have a clear picture of the environmental ramifications.

An analysis by Vliex et al. of rockets launched in 2022 revealed that spaceflight depletes the ozone layer and contributes to global warming, with a significant portion of this ozone loss attributable to nitrogen oxide emissions released by objects reentering Earth’s atmosphere.

The researchers calculated emissions from all 186 rockets launched in 2022, as well as all 472 objects—with a combined total mass of nearly 5,000 tons—that reentered the atmosphere that year. They conducted computational simulations of each launch’s trajectory and emissions at various altitudes up to 100 kilometers, and they calculated emissions released by object reentry. They also accounted for the effects of chemical reactions that occur in rocket exhaust plumes, which alter emissions’ chemical composition.

Incorporation of the calculated emissions into GEOS-Chem, a computational model of atmospheric chemistry, revealed their ozone-depleting and Earth-warming effects, with reentry emissions identified as playing a key role in ozone depletion. The researchers found that accounting for plume reactions reduced the estimated effects of spaceflight emissions, highlighting the value of considering plume chemistry in future assessments.

The analysis also underscored the varying effects of different rocket fuel types. Solid-state fuels, used recently in rocket boosters for NASA’s Artemis II mission to return astronauts to the Moon, appeared to cause the greatest amount of ozone depletion relative to propellant mass, while rocket-grade kerosene caused the greatest amount of warming.

On the basis of their findings, the researchers call for further research into reentry emissions and rocket plume chemistry as the space industry continues to expand and evolve. (Earth’s Future, https://doi.org/10.1029/2025EF007795, 2026)

—Sarah Stanley, Science Writer

Citation: Stanley, S. (2026), Rocket launches and reentries harm Earth’s ozone layer, Eos, 107, https://doi.org/10.1029/2026EO260183. Published on 8 June 2026.

Text © 2026. AGU. CC BY-NC-ND 3.0
Except where otherwise noted, images are subject to copyright. Any reuse without express permission from the copyright owner is prohibited.

Source: AGU Advances

The Sun continuously blasts charged, magnetic field–carrying particles, or plasma, in all directions. This solar wind interacts with the magnetic fields and atmospheres of several of our solar system’s planets and other bodies, sculpting long magnetic tails of charged particles—magnetotails—that stretch into space behind them.

Magnetotails contain thin layers of electric current–carrying plasma sheets, which sometimes “flap” in an up-and-down waving motion. Spacecraft observations have revealed that flapping in Earth’s magnetotail can be driven by a process called magnetic reconnection, in which magnetic field lines rapidly break and then snap together in a new configuration, releasing stored energy. However, whether reconnection plays this same role beyond Earth has thus far been a mystery.

Wen et al. report the first evidence that magnetic reconnection may also trigger magnetotail flapping at Mars.

Unlike Earth, Mars lost its global magnetic field billions of years ago. But it still sports a magnetotail, thanks in large part to interactions between the solar wind and charged particles in its upper atmosphere. Strong magnetic fields embedded in certain patches of the Martian crust—remnants of its lost planet-wide field—also influence the magnetotail.

Until recently, Mars’s magnetotail could only be studied using observations from NASA’s Mars Atmosphere and Volatile Evolution (MAVEN) spacecraft. MAVEN showed that the Martian magnetotail is highly dynamic, with a structure that twists, shifts, and flaps—and from which charged particles may escape into space. But because MAVEN can observe only one part of the magnetotail at a time, it couldn’t identify what processes might trigger flapping.

Another spacecraft, China’s Tianwen-1 orbiter, has now provided a second set of eyes. The researchers analyzed simultaneous observations from the two spacecraft, finding that signatures of magnetic reconnection detected by MAVEN in the upstream part of the magnetotail tended to coincide with flapping events detected downstream by Tianwen-1.

Before or during flapping, the spacecraft also detected temporary, twisted plasma structures known as flux ropes. A similar link has previously been observed on Earth, and it suggests that flux ropes generated by magnetic reconnection upstream might propagate downstream, driving instabilities in the magnetotail’s plasma sheets and triggering flapping.

Though more research is needed to confirm these findings, they shed new light on how energy moves and is released in space around Mars—and possibly other planets and celestial objects. (AGU Advances, https://doi.org/10.1029/2026AV002343, 2026)

—Sarah Stanley, Science Writer

Citation: Stanley, S. (2026), What makes Mars’s magnetotail flap?, Eos, 107, https://doi.org/10.1029/2026EO260123. Published on 20 April 2026.

Text © 2026. AGU. CC BY-NC-ND 3.0
Except where otherwise noted, images are subject to copyright. Any reuse without express permission from the copyright owner is prohibited.

The retreat of glaciers and ice sheets is expected to have widespread impacts on communities around the world because of its effect on sea levels. Already, the global average sea level is more than 10 centimeters higher than it was just 3 decades ago; and the rate of rise is increasing, contributing to increased storm surges and flooding, lost infrastructure and community lands, and more.

Recent reports on the instability of Antarctica’s Thwaites Glacier, for example, have focused attention on how accelerating ice flow can lead to ice sheet collapse and rising sea levels. Yet there is still substantial uncertainty about how quickly Thwaites and other glaciers will lose ice, in part because we don’t fully understand the myriad processes that contribute to their mass balance.

Earth’s ice sheets accumulate ice through snowfall and lose mass through a mix of surface ablation, iceberg calving, and melting at their interface with the ocean. Glacial ice flows under its own weight, and the rate at which it flows to coastal areas is a primary control on ice sheet mass loss.

Flow rates depend on how much resistance an ice sheet encounters at its interface with the ground (e.g., whether it is frozen to its substrate) and on its effective viscosity, a measure of how strongly it resists deformation. The viscosity of ice, in turn, varies based on properties including temperature, crystal size and orientation, and impurity content.

Some properties within and beneath ice sheets that affect how they flow are anisotropic, meaning they vary by direction. For example, roughness in some directions at the ice bed can facilitate ice sliding more effectively than roughness in other directions, similar to the way a properly oriented corrugated metal roof allows snow to slide off. Several forms of anisotropy within ice also affect how ice flows from land to ocean (Figure 1).

Measuring anisotropic properties is key to better understanding how quickly changes at the edges of the Greenland and Antarctic ice sheets will lead to sea level rise. Recent advances in ice-penetrating radar technology and in processing radar data are revolutionizing how we observe directionally varying ice sheet properties, paving the way for projections of mass changes that account for previously neglected processes.

Crystal Fabric: Memory and Modulator of Ice Flow

Fabric, the orientation of crystals composing ice, is the best studied and arguably most important of anisotropic ice sheet properties. As ice deforms, for example, by stretching horizontally as it flows toward the coast, its millimeter-scale crystals are reoriented (Figure 1).

Fabric thus contains a memory of past flow. Simultaneously, fabric influences flow because ice crystals are about 3 orders of magnitude easier to shear in some directions than others—similar to how stacked playing cards slide easily against each other when held along their edges but resist motion when pinched top to bottom.

The potential importance of fabric on large-scale ice flow has long been recognized, but a shortage of observations has made it difficult to quantify and validate its effect in ice sheet models. Until recently, fabric could be measured only directly in ice cores or inferred through seismic soundings. These methods provide highly detailed information about how fabric develops but are expensive, logistically taxing, and provide information only about sparse point locations.

Over the past 20 years, though, radar polarimetry has matured into a quicker and easier alternative means for inferring fabric, enabling observations at the scale of entire glaciers and providing new constraints on how fabric influences ice sheet flow.

How Radar Reveals Fabric

Ice-penetrating radar instruments emit electromagnetic energy as radio frequency waves. These waves reflect off interfaces within and beneath glacial ice, including transitions in ice chemistry and the contact surface between the ice sheet and the ground or water below. The properties of the reflected waves are then measured when they return to the radar. Just as fabric leads to anisotropic ice deformation, it also introduces directional dependence in the measured electrical properties.

The speed of a radar wave through an ice crystal is approximately 1% faster if the wave is polarized across the crystal’s principal (c) axis rather than aligned with it. Though small, this difference can compound enough that it causes measurable changes in returned radar signals.

In a typical radar survey over anisotropic ice, waves with different polarizations travel at slightly different speeds (Figure 2). The times that return signals arrive back at the receiver thus vary directionally, a difference that can be identified using polarimetric radars that transmit and receive radio waves at multiple orientations.

Fabric’s effect on radar signal travel times accumulates through an ice column, so it is more prominent in thicker ice with stronger horizontal fabric (i.e., the ice crystals are more consistently aligned). In such cases, differences in travel times between polarizations can be measured even by standard radars.

When fabric is weaker or ice is thinner, the offset is smaller and detectable only by systems that can identify the phases of radar returns—that is, the exact positions of the returned waves in their oscillation cycle. Even small wave speed differences from weak fabrics accumulate into measurable phase shifts between polarizations, which can be used to determine the consistency of crystal alignment and the predominant crystal orientation.

Small differences in fabric through an ice column can also change the strength, or amplitude, of returned signals. This amplitude difference offers an independent way to identify fabric orientation and its depth variation.

Polarimetric radar has been widely applied in cryospheric science in recent years largely due to the advent of low-cost systems that can measure signal phases. For example, the popular Autonomous phase-sensitive Radio Echo Sounder (ApRES) is a lightweight, ground-based system that can be used to infer ice fabric at single points down to 2 kilometers deep. In the past decade, polarimetric ApRES systems have revealed ice flow histories, including changes in flow directions, of key glaciers over the past few millennia. These measurements offer windows into how ice sheets responded to previous climate variations.

The next generation of polarimetric radars go beyond one-point-at-a-time stationary soundings, offering full polarimetry capabilities on moving platforms. These systems may soon allow scientists to map directional ice properties at the scale of entire ice sheets.

Insights into Fast-Flowing Ice Fabric

The growing number of radar studies conducted near sites where ice cores have been collected, which allow fabric to be investigated up close, has provided validation and bolstered confidence that fabric can be inferred accurately from its effects on radar. Researchers now infer fabric from radar in more dynamic areas, such as Thwaites Glacier, Whillans Ice Stream, and the Northeast Greenland Ice Stream (NEGIS), where ice fabrics change over short spatial scales and where drilling ice cores is logistically difficult. Airborne radar surveys are particularly effective in these settings because they can efficiently map fabric variations across large, fast-moving areas.

Observations of strong fabrics in fast-flowing regions suggest that fabric is an important control on ice viscosity, although its implications for ice flow are just beginning to be explored. For example, at Rutford Ice Stream in Antarctica, ApRES data indicate that fabric causes sharp changes in viscosity in different directions with depth, a complexity not captured by current ice flow models.

A combination of airborne and ground-based radar shows that the fabric of the NEGIS varies substantially across the ice stream, which facilitates horizontal shear that allows faster and more cohesive flow in the middle of the ice stream while simultaneously stiffening this ice against along-flow stretching. These viscosity variations may alter how quickly coastal changes, such as increased melt due to climate warming, influence inland ice flow.

The emerging consensus from radar observations and recent progress in fabric modeling is that ice fabric can soften ice stream shear margins by a factor of 10. In other words, the fabric tends to develop in a way that greatly reduces the ice’s effective viscosity at lateral boundaries between fast-flowing and slower-flowing ice, which enables the ice to deform more easily at the margins. The agreement between observations and process-scale modeling highlights fabric as a major, but largely ignored, control on ice flow that may affect estimates of how ice dynamics will contribute to future sea level rise.

Beyond Fabric

Most polarimetric radar studies so far have focused on fabric, but other ice characteristics can cause directional effects too. For instance, bubbles trapped in ice have dramatically different properties than ice itself. Ice deformation can bring bubbles into alignment, such that they affect radar waves differently in different directions.

Likewise, ice at its melting point can contain liquid water along boundaries between crystals, and if those pockets of water are aligned in one direction, they can also affect radar returns. Each of these properties has important influences on ice flow, but their implications are yet to be explored.

Another source of anisotropy is the bottom boundary of the ice sheet. This interface can be rougher in some directions than others, though the roughness is typically aligned with the prevailing ice flow direction or the direction of meltwater trapped within the ice.

Polarimetric radar can measure directionally dependent properties of ice sheet bases at a finer scale than radar profiling can. Such work is leading to new insights into glacier geomorphology, interactions of ice shelf bottoms with the underlying ocean, and how ice slides over substrate surfaces. Rates and extents of sub-ice-shelf melt and basal sliding are widely recognized as key controls on the future of the ice sheets.

Expanding Horizons: Large-Scale and Planetary Applications

Radar polarimetry has already transformed our understanding of ice fabric, revealing much about how crystal alignment modulates the flow of Earth’s ice sheets and filling critical gaps between the handful of direct measurements from ice cores. As polarimetric techniques mature, their applications are expanding.

Researchers are moving from studying isolated profiles of ice fabric to mapping it across whole basins, a key shift for validating bespoke models of fabric and its effects on flow. These models are also rapidly developing to include additional physical processes (e.g., migration recrystallization) and key simplifications (e.g., reducing directionally varying viscosity to a single number) that allow them to interface more easily with—and be incorporated into—large-scale models used for projecting sea level rise.

Techniques pioneered for measuring ice on Earth may also prove useful elsewhere in the solar system. Orbital radar sounders have already probed Mars’s ice masses, and the icy shell of Jupiter’s moon Europa will soon be surveyed by single-polarization radars aboard NASA’s Europa Clipper and the European Space Agency’s Jupiter Icy Moons Explorer (JUICE). These radars might be useful for polarimetry at some locations on Europa, which could reveal past and present motion of ice features and answer fundamental questions about the moon. Whether Europa’s shell flows, for example, may be key to whether its subsurface ocean can harbor life.

As polarimetric radar systems become routine tools for glaciologists and as similar instruments begin operating on spacecraft exploring icy worlds, a technique once limited to a few isolated core sites on Earth could be poised to transform our understanding of ice across the solar system.

Author Information

David Lilien (dlilien@iu.edu), Indiana University Bloomington; T. J. Young, University of St Andrews, Fife, Scotland; Benjamin Hills, Colorado School of Mines, Golden; Tamara Gerber, Université de Lausanne, Lausanne, Switzerland; and Matthew Siegfried, Colorado School of Mines, Golden

Citation: Lilien, D., T. J. Young, B. Hills, T. Gerber, and M. Siegfried (2026), New directions in mapping ice sheet fabrics and flow, Eos, 107, https://doi.org/10.1029/2026EO260154. Published on 14 May 2026.

Text © 2026. The authors. CC BY-NC-ND 3.0
Except where otherwise noted, images are subject to copyright. Any reuse without express permission from the copyright owner is prohibited.