Remote Sensing Labs
Browse all docs

Snow and Ice

Most surfaces are read by asking what they are. Snow and ice invert the usual difficulty. In open terrain they are among the easier things to find — few other materials announce their presence so plainly to a passing sensor — and among the hardest to describe once found. The brightness that makes a snowfield trivial to outline says almost nothing about the qualities that decide what the snow will do: how old it is, how wet, how deep, how much sunlight it still throws back. Observing the cryosphere is largely the work of getting past that first easy answer to the harder ones behind it.

The starting point is a spectral contrast: fresh snow is intensely bright where the eye sees and falls away sharply further into the infrared, the shape described under Spectral Bands. The standard way to exploit it is the Normalized Difference Snow Index, NDSI = (green − SWIR) / (green + SWIR). Snow scores high; a threshold near 0.4 is the long-standing convention for “snow”, but it is a starting point, not a constant — forests, shadow, and thin snow all shift where the right cut sits, which is why some products publish the NDSI value and leave the threshold to the user. What the contrast does not hand you is the rest: the same signature that locates snow is nearly mute about the snow’s condition, and the view that captures it is the one most often taken away.

Extent is one question; properties are several

The first thing a cryosphere observation can offer is extent — is there snow or ice here, or not — and on open, sunlit ground it offers it comparatively cleanly, so snow and sea-ice cover are among the more dependable things remote sensing reports. Even extent has its hard cases: snow beneath a forest canopy is largely hidden, and thin or patchy cover fills pixels only partly, which is why many snow products report a fraction or an index value rather than a yes/no. But extent is a single question, and almost everything a reader actually wants to know is a different kind of question about the same pixel. Grain size, which grows as snow ages and lowers its reflectance in the near- and shortwave-infrared; impurities such as dust and soot, which darken it in the visible and can hasten melt; liquid water content, which marks the difference between a frozen pack and a melting one; depth and the water stored in it, known as snow water equivalent; the albedo that governs how much sun the surface returns to the sky. None of these follows from presence, and each needs its own evidence. The same brightness that answers whether is close to silent on which and how much, and treating a confident extent map as though it had also settled properties is the most common way to over-read the cryosphere.

Why the optical view keeps going dark

The reflected-light view of snow and ice is not only partial in what it measures; it is often simply absent. Three conditions conspire, and all of them bear hardest exactly where the cryosphere lives. Polar winter removes the sunlight a reflectance depends on for weeks or months at a time, so the optical record over the highest latitudes has a season-shaped hole in it. Even when the sun is up it sits low, grazing the surface at an angle that lengthens shadows and complicates the geometry of what is measured. And cloud is persistent over the Arctic Ocean and the Southern Ocean, far more so than over the mid-latitude land where the optical habit was formed.

Cloud brings a second, sharper trouble, because snow and cloud look alike to any sensor tuned only to brightness — a confusion Cloud Masking treats from the cloud’s side, as contamination to be screened out. From the cryosphere’s side the consequence runs the other way: the clearest optical map of a snowfield is also the one most likely to be interrupted by the very cloud it can be mistaken for, and snow wrongly flagged as cloud is removed with it, so a run of scenes that each look clean can still under-report snow. The optical view of the cryosphere is a good view that is frequently unavailable, and its gaps are not evenly spread.

The view that survives cloud and night

When reflected light fails, an active microwave sensor is the natural second observer, because it carries its own illumination and its signal passes through cloud and darkness. SAR Basics explains why that is so, and why the resulting scenes must be read differently from a photograph; the point here is only what the cryosphere in particular gives such a sensor to say. Two distinctions stand out. The first is wet snow against dry: liquid water in the pack absorbs microwaves strongly, so backscatter drops sharply when melt begins — the opposite of the brightening wetter soil usually produces under SAR Basics — while dry snow is nearly transparent at common radar wavelengths. The onset of melt — the moment a snowpack begins to let water go — therefore appears as a darkening a reflectance would miss entirely, and it appears in the dark and under cloud, when it tends to matter most. The second is sea ice against open water, two surfaces whose microwave signals differ enough to be told apart when the optical scene is black or overcast, which is precisely the condition the polar ocean spends much of the year in. The long, near-daily record of sea-ice extent comes from passive microwave radiometers, which sense the natural microwave emission of ice and water through cloud and darkness at coarse resolution — tens of kilometers; radar adds the fine detail of floes and leads, though wind-roughened open water can mimic ice in a single radar scene. Neither replaces the optical map; each answers where the optical map cannot be made at all.

Coarse pixels and a polar clock

Two practical limits shape what any of this can resolve, and the cryosphere sharpens both. Resolution covers the general trade between how fine a pixel is and how much ground a sensor can cover; what the cryosphere adds is that some of the features that matter most are small against a coarse pixel — a lead of open water threading through sea ice, the ragged edge where a melt is advancing — so a sensor built for broad coverage averages across exactly the boundary a reader is watching. Timing bites in its own way. Time Series develops the general problem of spacing and gaps; here the particular difficulty is that the cryosphere changes fast and the optical cadence is thinnest just when change is quickest — a melt onset can turn a surface over in days, while the polar season is starving the optical record of usable passes at the same moment. A snapshot that arrives a week late over a melting margin is not a slightly worse measurement of the same thing; it is a measurement of a surface that has already moved on.

Trusting a snow or ice number

Almost every quantity worth having here — snow-covered area, ice extent, the presence of melt, an albedo — is retrieved rather than directly measured: derived from what the sensor saw under stated assumptions, not read off a dial. That makes disclosure the thing such a product owes its reader. Which sensing mode produced it matters, because an optical extent and a microwave melt signal fail in different ways and on different days. Some confusions are structural and worth naming plainly: in radar, wet snow’s dark return can resemble open water or smooth bare ground, and a surface answer can be quietly stood in for by what lies just beneath it, since neither light nor microwaves report only the skin they appear to. A number that does not carry its mode, its date, and the conditions it was retrieved under cannot be trusted at a single point, however plausible it looks. Comparing a place against its own history is the firmer use. How a trend is built from an archive, and what it can be held to, is covered under Climate.

Sources