Satellites can scan the Arctic again and again, which a field team on the ice cannot do. But a satellite sensor does not receive a label such as “first-year ice” or “1.4 metres thick.” It records radiation emitted or reflected by the surface, or a radar pulse that returns to the antenna. Scientists need measurements from the ice to work out how that signal relates to ice properties.

At a drill hole, researchers can measure thickness directly. Ice cores and snow probes can measure salinity and snow depth at specific locations. A satellite covers a much larger area, but its signal has to be converted into estimates of thickness, snow depth, roughness, or ice type. Measurements collected on the ice provide known values at particular places and times, so researchers can build that conversion and test it elsewhere.

What field measurements are for

Suppose a satellite product reports ice thickness. Researchers need to check both how the satellite signal is converted to thickness and how well the resulting estimate works:

• Calibration builds or adjusts the conversion from the satellite signal to thickness. Researchers may compare radar returns with thickness, snow depth, roughness, temperature, and salinity measured at the same sites.
• Validation tests the finished conversion against independent observations. The measurements used to tune an algorithm cannot also serve as an independent test of it.

These direct observations are often called ground truth, but they still contain uncertainty. A drill measurement may be slightly wrong, and even an accurately measured hole may not represent the much larger satellite pixel around it.

Field instruments can also record changes between satellite passes. Instruments left on a floe show when ice starts to grow or melt, how snow changes the temperature gradient, and how ocean heat affects the underside of the ice. Ice cores show how salinity changes from top to bottom. Cameras and light sensors below the ice record algae and the light available to them. At the same site, researchers can then compare atmospheric cooling and snow insulation with ice growth, ocean heat at the lower boundary, and changes in under-ice habitat across the Arctic's coupled atmosphere, snow, ice, ocean, and biology.

Working on moving ice

A floe is a piece of floating sea ice. A camp built on one moves with the ice rather than staying at one geographic location. By drifting with the same floe, researchers can follow its freeze-up, winter growth, deformation, and melt.

A field programme may use several platforms at once:

• a drifting camp puts people and instruments directly on the floe;
• an icebreaker carries people and supplies and provides laboratories, power, and a safer base beside the work area;
• buoys stay behind when the team leaves and transmit measurements as the ice drifts;
• balloons, ocean instruments, and remotely operated vehicles collect measurements above and below the ice.

During the 2019–2020 MOSAiC expedition, the icebreaker Polarstern sat at the centre of a drifting observatory. Researchers set up a camp on the ice and a wider network around it. They measured the atmosphere, snow, ice, ocean, and ecosystem through the seasons. Because those records covered the same floe and period, they could compare changes in ice growth or melt with the weather, snow cover, ocean heat, and ecosystem conditions around the ice.[2]

The floe can move enough to break up the experiment itself. When the ice diverges, a lead—a crack or channel of open water—can open between the ship and an instrument site. When it converges, broken ice can pile into a pressure ridge across a transect or cable. Cold, storms, darkness, poor visibility, and wildlife hazards also limit when people can work. Ice thickness varies sharply over short distances, so no single thickness makes travel safe in every place and season. Route decisions also depend on the ice type, recent weather, currents, cracks, snow cover, load, and local knowledge.

Why one hole cannot represent a whole pixel

A drill hole may be only $5\,\mathrm{cm}$ across, while a passive-microwave pixel may be about $25\,\mathrm{km}$ across. Snow-probe lines, electromagnetic surveys, aircraft, and finer-resolution radar observations cover scales between those two.

From a drill hole to a satellite pixel

• Field measurement
• Satellite pixel

The survey lengths in the diagram are examples, not fixed limits for the instruments. A hole measures one spot, a transect samples points along a line, and a satellite pixel combines the signal from an area that may contain smooth ice, ridges, snowdrifts, melt ponds, and leads.

A dense grid of holes across one pixel would take far too long to drill. For a $25\,\mathrm{km}\times25\,\mathrm{km}$ pixel, one hole every $100\,\mathrm{m}$ gives 250 sampling positions along each side:

$$N=\left(\frac{25{,}000}{100}\right)^2=62{,}500\text{ holes}.$$

At ten minutes per hole and eight working hours per day,

$$\text{working days} =\frac{62{,}500\times10}{8\times60} \approx1302.$$

That is about 3.6 years of drilling every day, before allowing any time for travel, weather, rest, or equipment delays. The ice would drift and change long before the grid was finished.

Since a dense grid is impossible, field teams place fewer measurements where they will capture the main types of surface in the pixel. A transect is a line along which measurements are repeated. A line that crosses smooth first-year ice, rough ice, ridges, and snowdrifts captures variation that would be missed by a cluster of holes drilled only where access is easy.

Buoys add measurements through time, but they do not give dense coverage across the Arctic. If $N$ buoys were spread evenly across area $A$, their rough spacing would be

$$d\approx\sqrt{\frac{A}{N}}.$$

Across a winter ice area of $14\text{ million km}^2$, 100 buoys would be about 374 km apart; even 1000 would be about 118 km apart.

More buoys still leave wide gaps

This calculation assumes perfectly even spacing, so it is not a map of a real buoy network. A buoy can record one drifting location repeatedly, while satellites fill in the large areas between buoys. Neither provides the other kind of coverage.

Measuring thickness and snow

Drill holes provide direct thickness at individual points, an electromagnetic sled covers a transect quickly, and a snow probe measures the snow that must be removed from the sled's total distance.

Start with a direct thickness reference

An auger cuts a hole through the ice. Researchers lower a weighted tape with a hinged bar through the hole, then pull it up until the bar catches the underside of the ice. The distance from the top of the ice to the bar is the ice thickness. At the same site, they can measure snow depth and freeboard, the height of the ice surface above the water.

The tape reaches the ice bottom directly, which makes it a useful reference for other instruments. But it covers only one point, so a few accurate holes do not show how thickness varies between them.

Extend the line with electromagnetics

An electromagnetic (EM) instrument uses the strong conductivity contrast between relatively resistive sea ice and conductive salty seawater. Its transmitter creates a changing magnetic field. This induces electrical currents in the seawater, and those currents create a secondary magnetic field measured by the receiver. Nearby seawater produces a stronger response than seawater farther below under comparable conditions. An EM sensor pulled on a sled can therefore estimate the distance from the sensor to the water along a transect.

An EM sled senses how far away the conductive seawater is

Snow depth is measured separately before the sensor-to-water distance is converted to sea-ice thickness.

The distance from the sensor to the water is not ice thickness on its own. It includes the instrument's height above the surface and, when the sled crosses snow, the combined depth of snow and ice. Researchers subtract the instrument height and snow depth to estimate the ice thickness. If $30\,\mathrm{cm}$ of snow is left in the total, the ice thickness will be overestimated by about $30\,\mathrm{cm}$.

A snow probe measures that upper layer. It is pushed through the snow until it reaches the harder ice surface, while a logger records depth and position. Wind removes snow from smooth areas and piles it around rough ice and ridges, so researchers take probe measurements all along the transect rather than only beside a drill hole.

Snow also presses the floe farther into the water, changing the conversion from satellite freeboard to thickness. In the approximate sensitivity example from Reading Sea Ice from Space, a $10\,\mathrm{cm}$ error in snow depth produces about a $30\,\mathrm{cm}$ absolute error in inferred ice thickness. That result depends on the densities assumed in the example; it is not a universal multiplier or a relative-error percentage.

Retrieving snow depth from space is difficult, but satellite snow-depth products do exist. ESA introduced an operational product that combines CryoSat radar with ICESat-2 laser measurements in 2025. Researchers still compare its estimates with field measurements because snow depth, density, grain structure, wetness, and redistribution all change the signal used in the retrieval.[1]

Reading a temperature profile

A thermistor string is a vertical chain of temperature sensors installed through the snow, ice, and upper ocean. Water around the string freezes it into place, and it keeps sending profiles after the field team leaves. As freezing or melting changes the snow and ice layers, the temperatures and the depths where the profile changes slope also shift.

Consider an idealized midwinter case. The snow surface is at $-25^\circ\mathrm{C}$, the water below the ice is at its local freezing temperature of $-1.8^\circ\mathrm{C}$, the snow is $0.10\,\mathrm{m}$ thick, and the ice is $1.50\,\mathrm{m}$ thick.

A midwinter temperature profile has a kink at the snow-ice boundary

Read the graph downward from the snow surface:

1. Temperature rises quickly through the $0.10\,\mathrm{m}$ snow layer.
2. At the boundary between snow and ice, the line changes slope because the two materials conduct heat at different rates.
3. Through the $1.50\,\mathrm{m}$ ice layer, temperature rises more gradually per metre until it reaches $-1.8^\circ\mathrm{C}$ at the ice bottom.

The bottom is $1.60\,\mathrm{m}$ below the snow surface, but the ice thickness is $1.50\,\mathrm{m}$. The upper $0.10\,\mathrm{m}$ is snow.

The change in slope comes from thermal resistance. In a steady, one-dimensional winter model, the same conductive heat flux passes through the snow and ice. Snow conducts heat less readily than ice, so a larger temperature change per metre is needed to move the same heat through it. With thermal conductivities

$$k_s=0.3\,\mathrm{W\,m^{-1}\,K^{-1}} \qquad\text{and}\qquad k_i=2.0\,\mathrm{W\,m^{-1}\,K^{-1}},$$

the layer resistances are

$$R_s=\frac{0.10}{0.3}=0.333\,\mathrm{m^2\,K\,W^{-1}} \qquad\text{and}\qquad R_i=\frac{1.50}{2.0}=0.750\,\mathrm{m^2\,K\,W^{-1}}.$$

The total temperature difference is $-1.8-(-25)=23.2^\circ\mathrm{C}$. The temperature difference across the snow is therefore

$$\Delta T_s =23.2\left(\frac{R_s}{R_s+R_i}\right) \approx7.1^\circ\mathrm{C}.$$

Starting at the snow surface, the temperature at the snow–ice boundary is $-25+7.1\approx-17.9^\circ\mathrm{C}$. Just 10 cm of snow accounts for about $7.1^\circ\mathrm{C}$ of the total difference.

The axes can make the slope look backward at first. Temperature is horizontal and depth is vertical, so the larger temperature change per metre in snow looks flatter on the page, not more vertical.

The two straight line segments are an idealization. In real snow and ice, the profile changes with air temperature, snow redistribution, flooding, sunlight, melt, and ocean heat. The spacing between sensors also limits how precisely the layer boundaries can be located.

Measuring salt, albedo, and heat exchange

Researchers use an ice core to measure properties that an EM sled or satellite does not resolve from top to bottom. They cut the core into short sections, melt each section separately, and use the meltwater's electrical conductivity to estimate salinity. Keeping the sections separate gives a vertical salinity profile instead of one average for the whole core. They also measure temperature along the fresh core before it warms. Temperature and salinity together help them estimate the volume of brine in the growing sea ice, which affects ice strength and radar response.

Radiometers measure how much sunlight a surface reflects. An upward-looking radiometer records incoming shortwave radiation, while a downward-looking one records radiation reflected by snow, bare ice, or a melt pond. Taking both readings at the same time gives

$$\alpha=\frac{\text{reflected shortwave radiation}}{\text{incoming shortwave radiation}}.$$

The two measurements have the same units, so albedo $\alpha$ has no units. Repeating them over different surfaces shows how each surface contributes to the albedo of a satellite pixel that contains a mixture of snow, ice, ponds, and water.

A flux tower measures more of the surface energy budget. Its radiometers record incoming and outgoing shortwave and longwave radiation. Turbulence instruments estimate sensible- and latent-heat exchange with the atmosphere. Temperature profiles constrain the heat conducted through the ice, while instruments below the ice measure heat supplied by the ocean. Together, these measurements supply the terms in the surface energy balance. Comparing the terms shows which energy source supplied the heat when the ice melted.

A remotely operated vehicle can carry an upward-looking camera and light sensors beneath ice with different snow depths. The light sensors measure how much sunlight passes through the snow and ice, while the camera records algae growing within or beneath the ice. Surface images alone cannot show this under-ice habitat.

Connecting field measurements to a satellite map

A field campaign can nest measurements so that each larger survey crosses an area measured in more detail:

A campaign might use a 100 m drill line, a 1 km sled transect, a 10 km airborne survey, and a 25 km pixel. These are example survey lengths, not fixed resolutions for every instrument.

The measurements must overlap in time as well as space. A floe drifts, leads open, and wind redistributes snow between a field visit and a satellite pass. Teams record positions and times, track ice motion, and collect measurements as close to the overpass as practical. Even a perfectly timed survey will not represent the pixel if it samples only smooth ice while the rest of the pixel is dominated by ridges.

Calibration and validation use this nested survey in different ways. Measurements used to tune a relationship between radar and ice properties are calibration data. Researchers then apply the relationship elsewhere and compare the result with field measurements that were withheld from calibration. To validate a snow-depth product, they would:

• measure snow depth along representative transects and measure snow density if the retrieval requires it;
• sample smooth and rough ice, ridges, and drifts rather than following only an easy route;
• match the satellite footprint and overpass while accounting for ice drift;
• spread observations across the pixel instead of clustering many readings in one patch;
• keep an independent test set instead of judging the retrieval with its calibration data.

Satellite radar is also useful before the field team travels. It can show candidate floes, ridges, leads, and places where a retrieval looks unusual. Researchers can then visit selected sites, measure their physical properties, and check why the radar image looks the way it does.

Inuit knowledge and community observations

A satellite pixel may estimate ice concentration or roughness, but it cannot tell whether a particular route is safe today. Inuit knowledge of sea ice includes local currents, weather, ice types, travel conditions, freeze-up and breakup, and changes seen across generations. In Nunavut, Inuit Qaujimajatuqangit is a broader body of knowledge and ways of knowing; sea-ice knowledge is one part of it.

Treating this knowledge only as extra “ground truth” for an outside project would leave out how it is used. A route has to work for a particular traveller, vehicle, and load. Rough ice may support a load but be impossible to cross, while smooth new ice may be easier to travel over but still be dangerous. Thickness is only one sign of the ice's condition. Neither a satellite image nor one drilled value can guarantee that a route is safe.

SIKU is a platform developed by and for Indigenous communities. It brings community observations together with satellite imagery and other information, supporting community-led monitoring and travel decisions while letting people compare local observations with the regional view from satellites.[3]

What remains difficult

The conditions that most affect sea-ice measurements are often the hardest to sample. Snow changes quickly and unevenly. Thin ice may be too dangerous to approach. Leads and ridges can rearrange a survey site, and wet snow or melt ponds make both field access and satellite retrievals harder. A campaign lasts for a limited time, but the ice keeps changing all year.

No instrument can cover every place, scale, and season. Teams therefore combine direct point measurements, transects, drifting records, aircraft, and satellites, then match them as closely as possible in location and time. Snow transects are especially important for satellite thickness estimates: snow pushes the floe down and changes the conversion from freeboard to total thickness, so an error in snow depth becomes an error in inferred ice thickness.

Sources

[1] https://earth.esa.int/eogateway/news/snow-depth-on-sea-ice-among-new-and-improved-cryosat-data-products [2] https://mosaic-expedition.org/expedition/drift [3] https://www.asc-csa.gc.ca/eng/blog/2025/09/29/siku-traditional-knowledge-meets-satellite-data.asp