1. Why Snow and Clouds Look Alike in an Ordinary Satellite Photo
NASA's Earth Observatory shows a scene of the north-central United States taken on January 2, 2001, by an instrument called MODIS on NASA's Terra satellite. In visible light (red light at about 0.66 micrometers, written 0.66 µm), snow is about as bright as clouds, so the two are hard to tell apart[1]. NOAA's snow-mapping center reports the same trouble with the visible channels on its own weather satellites[2].
That matters because a snow map is only useful if it is right. A cloud mistaken for snow makes the snowy area look bigger than it is. Meanwhile, ground hidden under a cloud cannot be seen at all, and a good map has to admit that.
2. Why Snow Looks Dark and Clouds Look Bright at 1.6 Micrometers
Sunlight contains more than the colors we see. Just past red light lies near-infrared light, and at about 1.6 µm snow behaves very differently from a cloud. NASA explains that snow absorbs sunlight at this wavelength and looks much darker than clouds, which lets MODIS separate the two very effectively[1]. NOAA's snow-mapping center describes the same thing: at 1.6 µm snow has very low reflectance, while the reflectance of clouds stays high. It adds that both thin, high cirrus clouds and thick clouds can be told apart from snow at this wavelength[2].
So for one patch of ground, the satellite has two measurements. Snow is bright in visible light and dark at 1.6 µm. A cloud is bright at both. Those two different patterns are the clue.
The sources used here do not explain why snow absorbs this wavelength so strongly. Ice absorbing it is the natural guess, but that is a guess, so this article leaves it there. It is a good question to carry to a physics textbook.
3. How a Computer Turns That Difference into a Snow Index
A computer cannot judge "dark" and "bright" by eye, so the difference is turned into a number. The Normalized Difference Snow Index (NDSI) is calculated as (green − shortwave infrared) ÷ (green + shortwave infrared). For MODIS, green light is band 4 and the shortwave infrared band is band 6[3].
To see how the formula behaves, here is a made-up example. These are not real measurements. Suppose a snowy pixel reflects 0.8 of the green light and 0.1 of the 1.6 µm light. Then NDSI = (0.8 − 0.1) ÷ (0.8 + 0.1), which is about 0.78. A cloudy pixel might reflect 0.8 green and 0.6 infrared: (0.8 − 0.6) ÷ (0.8 + 0.6), about 0.14. Snow ends up with a much bigger number.
A number alone is not the whole decision, though. In the MODIS method described by EUMeTrain, a European training service for weather forecasters, a pixel is mapped as snow if its NDSI is at least 0.4 and its reflectance in band 2 is above 11%. It is not mapped as snow if its band 4 reflectance is under 10%, which keeps very dark pixels out. The method uses a pixel only if it is daytime, the pixel is over land and not covered by cloud, and its brightness temperature (a temperature worked out from the infrared light the satellite measures) is below 283 K (about 10 °C)[4]. The 0.4 cutoff is one published example, and other products and satellites may use different values.
4. How the Hidden Light Becomes a Color Picture You Can Read
People also need to see the difference. One way is a false-color image, where each color channel is fed by a different wavelength. NASA's picture of the Midwest is one: land is green, water is black, snow is red, and clouds are white, while clouds with plenty of ice particles look pinkish[1]. Because snow is dark at 1.6 µm, that wavelength's missing light is what lets snow stand out from the white clouds.
Japan's weather agency does something similar with its Himawari-8 and Himawari-9 satellites. (Himawari means "sunflower" in Japanese.) Its "Natural color RGB" image uses the 1.6 µm band as red, a 0.86 µm near-infrared band as green, and the visible 0.64 µm band as blue. It helps forecasters tell ice clouds from water clouds and pick out snow and ice, vegetation, desert dust and volcanic ash. It works only in daylight[5]. The agency's page does not say what color snow ends up, so this article does not guess.
5. When Satellites Still Confuse Snow and Cloud
None of this is perfect. According to the EUMeTrain course, the most frequent errors in MODIS snow maps come from the difficulty of telling snow from cloud, and very thin snow, under 1 cm, is also hard to detect. When the maps were checked against ground data, the overall accuracy of the daily product was reported as 93%, but it varies with the type of land and the snow itself[4]. So the number is a good sign, not a promise for every place.
That is why snow maps are usually built from many days of images and are marked wherever clouds block the view. A pixel hidden by cloud is better left unknown than guessed.
6. How You Can Look at Snow and Cloud Through a Satellite's Eyes
You can compare the two kinds of picture yourself. Open NASA's Midwest image[1] and notice that the same scene, shown in a way that uses 1.6 µm light, separates the red snow from the white clouds. If it snows where you live and a satellite image is available, look at a picture of your area after a snowfall and ask which of the white patches could be cloud. When snow is forecast, you could compare a later picture with the first one, but remember that clouds move and new snow can fall.
Questions the sources here leave open: why does ice absorb 1.6 µm light so strongly, and how well can a satellite tell a snowfield from a cloud made of ice crystals? Papers by the MODIS snow-mapping team are a good place to start looking.