Jeff Dozier
NASA Goddard Space Flight Center
Greenbelt MD 20771


Mapping of snow and estimation of snow characteristics from satellite remote sensing data require that we distinguish snow from other surface cover and from clouds and compensate for the effects of the atmosphere and rugged terrain. The spectral signature of the snowpack is calculated from a radiative transfer model, accounting for scattering and absorption by the ice grains, water inclusions, and particulates. Large surface grain sizes can be distinguished from areas where the grain size is finer at the snow surface, using Landsat Thematic Mapper bands 2 (green), 4 (near-infrared), and 5 (short-wave infrared). Because of saturation in band 1 (blue), estimation of the degree of contamination by absorbing aerosols is not feasible.


Water in its frozen states accounts for more than 80 per cent of the total fresh water on Earth and is the largest contributor to rivers and ground water over major portions of the middle and high latitudes. Snow and ice also play important interactive roles in the Earth's radiation balance, because snow has a higher albedo than any other natural surface. Over 30 percent of the Earth's land surface is seasonally covered by snow, and 10 percent is permanently covered by glaciers. Snow cover represents a changing atmospheric output resulting from variability in the Earth's climate, and it is also a changing boundary condition in climate models. Thus understanding of global and regional climates and assessment of water resources require that we monitor the temporal and spatial variability of the snow cover over land areas, from the scale of small drainage basins to continents.

The alpine snow cover and alpine glaciers in the mid-latitudes are important in both the climatic and hydrologic contexts--to examination of global and regional climates and to use of water resources (Colbeck et al., 1979; Walsh, 1984). Much of the uncertainty and sensitivity in the global hydrologic cycle lies in these reservoirs of frozen water, and the melting of alpine glaciers during the last half-century appears to account for much of the corresponding rise in sea level (Meier, 1984).

Over the last two decades satellite remote sensing has opened the possibility of data acquisition at regular intervals, and operational as well as research-oriented satellites have provided information on snow cover. Remote sensing of the seasonal snow cover has been used to improve the monitoring of existing conditions and has been incorporated into several runoff forecasting and management systems. Satellite sensors in the visible and near-infrared wavelengths provide information on the spatial distribution of parameters of hydrologic importance. From these reflectance measurements, we can measure snow-covered area, rates of snow-cover depletion, and surface albedo. The Landsat systems, in particular, are a source of data for hydrological and glaciological research at the scale of drainage basins.


The scattering and absorption of light by the snowpack, clouds, and the atmosphere are analyzed with a multiple-scattering model, a two-stream approximation to the radiative transfer equation (Meador and Weaver, 1980). The fundamental scattering properties of the ice grains and water inclusions in the snowpack, the water droplets or ice crystals in clouds, and the aerosols in the atmosphere are calculated by the complex angular momentum approximation to the Mie equations (Nussenzveig and Wiscombe, 1980). The LOWTRAN model (Kneizys et al., 1988) is used to obtain values for molecular absorption in the atmosphere at the desired wavelengths.


Snow is a collection of ice grains and air, and, when at 0 degrees C, it also has a significant fraction of liquid water. Snow also often includes particulate and chemical impurities - dust, soot, pollen and other plant material, and small amounts of the major cations and anions. Thus the optical properties of snow depend on the bulk optical properties and the geometry of the ice grains, the liquid water inclusions, and the solid and soluble impurities. Similarly, clouds are composed of water droplets, sometimes ice crystals, and they may contain impurities.

Bulk Optical Properties of Ice and Water

In the visible and near-infrared wavelengths the bulk optical properties of ice and water are very similar, so the reflectance and transmittance of the snowpack in this region of the electromagnetic spectrum depend on the wavelength variation of the refractive index of ice, the grain size distribution of the snow, the depth and density of the snowpack, and the size and amount of those impurities whose refractive indices are substantially different from those of ice and water. The reflectance of wet snow in the near-infrared is lower than that of dry snow, but mainly because of microstructural changes caused by the water, except in some narrow spectral regions where the optical properties of water are different than those of ice. Similarly, the reflectance and transmittance of clouds depend on the geometric thickness, the number density of the droplets, and their size distribution.

The most important optical property of ice and water, which causes spectral variation in the reflectance of snow and clouds in visible and near-infrared wavelengths, is that the absorption coefficient (i.e. the imaginary part of the refractive index) varies by seven orders of magnitude in the wavelengths from 0.4 to 2.5 ~m. Figure 1 shows the complex refractive index n + i k for ice and water. The important properties to note are:

1. the spectral variation in the real part n is small, and the difference between ice and water is not significant;
2. the absorption coefficients k of ice and water are very similar, except for the region between 1.55 and 1.75 um, where ice is slightly more absorptive;
3. in the visible wavelengths both ice and water are highly transparent so k is small;
4. in the near-infrared wavelengths ice and water are moderately absorptive, and the absorption increases with wavelength.

Spectral Reflectance of Snow

The spectral and angular variation in snow reflectance are modeled by the radiative transfer equation, as shown to be appropriate by Bohren and Barkstrom (1974) and Warren (1982). In the visible wavelengths ice is highly transparent, so the albedo of snow is sensitive to small amounts of absorbing impurities (Warren and Wiscombe, 1980). In the near-infrared wavelengths ice is more absorptive, so the albedo depends primarily on grain size (Wiscombe and Warren, 1980). We make the following assumptions in modeling the reflectance of snow. Most of these assumptions have yet to be tested by rigorous measurements of physical properties and spectral reflectance of the same snowpack, but the model produces reflectance spectra that match those of snow (Warren, 1982).

1. The reflectance of snow is modeled as a multiple scattering problem. Scattering properties of irregularly-shaped grains are mimicked by Mie calculations for an "equivalent sphere," for which the best candidate in the wavelength region from 0.4 to 1.1 um is apparently the sphere with the same surface-to-volume ratio (Dozier et al., 1988), which can be measured by stereological methods applied to snow samples (Davis and Dozier, 1989). Although snow grains are irregularly shaped, they are usually not oriented, so the assumption that their scattering properties can be mimicked by some spherical radius r is reasonable, especially when we want to describe the general spectral properties. When we want details about the angular characteristics of the reflectance, the spherical assumption could become more critical.

2. Near-field effects are assumed unimportant. The fact that the ice grains in a snowpack touch each other apparently does not affect the snow's reflectance, because the center-to center spacing is still much larger than the wavelength. That is, snow reflectance is independent of density up to about 650 kg m-3. Reflectance measurements carried out under field conditions over a season and simply analyzed statistically will show a significant inverse relationship between density and reflectance, but the physical model shows that the explanation for changes in reflectance lies in other properties of the snow cover, namely an increase in grain size and in the amount of contaminants near the surface.

3. The effect of absorbing impurities (dust, soot) can be modeled either as separate spheres (smallest effect) or as concentric spheres with the impurity in the center (largest effect). These should bound the magnitude.

Figure 2 shows the spectral reflectance of pure, deep snow for snow grain radii from 50 to 1000 um (0.05 to 1.0 mm), representing a range from new snow to spring snow, although the grain clusters in coarse spring snow can exceed 5 mm in radius. Because ice is so transparent in the visible wavelengths, increasing the grain size does not appreciably affect the reflectance. The probability that a photon will be absorbed, once it enters an ice grain, is small, and that probability is not increased very much if the ice grain is larger. In the near-infrared, however, ice is moderately absorptive. Therefore, the reflectance is sensitive to grain size, and the sensitivity is greatest at wavelengths from 1.0 to 1.3 um. Because the ice grains are strongly forward-scattering in the near-infrared, reflectance increases with illumination angle, especially for larger grains, as shown in Figure 2.

Because the complex indices of refraction of ice and water are similar, liquid water per se has little effect on the reflectance of snow. Except where meltwater ponds in depressions when melting snow overlies an impermeable substrate, or when rain falls on fine-grained snow, liquid water content in snow rarely exceeds 5 or 6%. This small amount of water does not affect the bulk radiative transfer properties, except possibly in a few narrow wavelength regions where the absorption coefficients are appreciably different (Hyvarinen and Lammasniemi, 1987). Instead, the decreases in reflectance that occur as snow melts result from the effective size-increase caused by the two- to four-grain clusters that form in wet, unsaturated snow (Colbeck, 1979, 1986). These apparently behave optically as single grains.

Although the reflectance in the visible wavelengths is insensitive to grain size, it is affected by two variables, finite depth and the presence of absorbing impurities. Figure 3 shows spectral reflectance for a range of grain sizes of snow water equivalences from 10 to 100 mm, over a black surface. For large grains, r = 1 mm, reflectance of snow with a water equivalence of as large as 100 mm is less than that of a deep snowpack. In a similar manner, minute amounts of absorbing impurities reduce snow reflectance in the visible wavelengths (Warren and Wiscombe, 1980). Soot concentrations as low as 0.1 ppmw (parts per million by weight) are enough to perceptibly reduce reflectance. The effect of the absorbing impurities is apparently enhanced when they are inside the snow grains, because refraction focuses the light on the absorbers (Grenfell et al., 1981; Chylek et al., 1983; Bohren, 1986).

Spectral Reflectance of Clouds and Snow/Cloud Discrimination

In visible satellite data, clouds can usually be distinguished from snow by texture, but not when both snow and clouds saturate the sensor, as might be the case in the spring. Moreover, in a computer image-processing system, texture is more difficult to analyze than spectral information. Hence we seek wavelength bands where snow and clouds have different spectral signatures. Clouds may be either warmer or colder than the snow surface, so one cannot reliably distinguish clouds from snow in the thermal wavelengths. Properties that cause clouds to have different spectral reflectance than snow are, in order of importance:

1. Cloud droplets or ice crystals are smaller than snow grains. Cloud droplets usually have size radii less than 10 um; crystals in cirrus clouds can be as large as 40 um, but most of them are smaller. A smaller scattering element-droplet, crystal, or grain-is likely to absorb less radiation, but the difference is greatest at wavelengths where the medium is modestly absorptive.

2. Most clouds are composed of water droplets, even at temperatures below 0 degrees C. At most wavelengths in the optical region water and ice have similar refractive indices, but ice is slightly more absorptive from 1.55 to 1.7 um. The difference in the size of the scatterers between clouds and snow, however, is more important than the difference in composition.

3. Snow on the ground is usually optically thicker than clouds. Therefore in the visible wavelengths snow is sometimes brighter, because some of the light incident on the cloud is transmitted through it. Thick clouds, however, are as bright as snow, so they cannot be dependably distinguished in this wavelength region by a lower reflectance. Cirrus clouds are usually thinner and have lower amounts of water per column of unit cross-sectional area.

Figure 4 shows spectral reflectances for water and ice clouds. The water clouds used in the calculation have a water equivalent thickness of 10 mm, while the cirrus clouds have 1 mm. Therefore the water clouds are brighter. In the visible wavelengths the water clouds and the snowpack are of comparable reflectance.


The Thematic Mapper first orbited on Landsat-4, launched in 1982. Landsat-5 was launched in 1985, and its Thematic Mapper is still functioning, although it was designed only for a five-year life. Orbital altitude is 705 km, spatial resolution at the surface is 30 m, and the orbit has a 1 day repeat cycle. Swath width is 185 km. There are seven spectral bands: three in the visible (1, 2, and 3); one in the near-infrared (4); two in the short-wave infrared (5 and 7); and one in the thermal infrared (6). Table 1 shows the wavelen~ths and radiometric characteristics.

Table 1. Landsat-5 TM Radiometric Characteristics

(data from Markham and Baricer, 1987)

TM band 1 will usually saturate over snow, except in the shadows, during all months. TM2 and TM3 will not usually saturate in December or January, will occasionally saturate in February, and will frequently saturate on slopes exposed to the Sun throughout the spring. TM4 will only occasionally saturate, after new snow in the spring, and TM5 and TM7 should never saturate over snow in the mid-latitudes, but may saturate over clouds or bright soils.


Satellite remote sensing in the visible and near-infrared wavelengths has become increasingly important to snow hydrologists because the data provide information on the spatial distribution of parameters of hydrologic importance. In snow and ice studies, remote sensing has been used
to improve the monitoring of existing conditions and has been incorporated into several runoff forecasting and management systems. The principal operational use of remote sensing of snow properties has been to map the extent of the snow cover. Throughout the world, in both small and large basins, maps of the snow cover throughout the snow season are used to forecast melt, both in areas with excellent ancillary data and in remote areas with no ancillary data (Rango et al., 1977; Andersen, 1982; Martinec and Rango, 1986).

Since the first mapping of snow cover from satellite, the spectral and spatial resolution of the available sensors has been much improved. The high spatial resolution satellites such as Landsat and SPOT and the medium resolution sensors such as the NOAA AVHRR are widely used for mapping snow cover. The selection of the appropriate sensor depends on a tradeoff between spatial and temporal resolution (Rott, 1987). The Landsat Multispectral Scanning System (MSS) has a spatial resolution of about 80 m and is suitable for snow mapping in basins larger than about 10km2 (Rango et al., 1983). Improved spatial resolution has been available since 1982 from the Landsat Thematic Mapper (30m) and since 1984 from the French SPOT satellite (20m in the multispectral mode and 10 m in the panchromatic mode). SPOT has the finest spatial resolution, but the Thematic Mapper has the best spectral coverage. Dozier (1989) explains how snow cover and snow properties can be mapped from the TM.

Incorporation of Topographic Effects

In all but very gentle terrain, significant variation in remotely sensed images in visible and near infrared wavelengths results from local topographic effects that cause variation in illumination angle and shadowing from local horizons (Williams et al., 1972; Dozier and Outcalt, 1979; Olyphant, 1984; Dozier and Frew, 1990).

In the solar spectrum, irradiance in alpine terrain has three sources: (1) direct irradiance from the sun; (2) diffuse irradiance from the sky, where a portion of the overlying hemisphere is obscured by terrain; and (3) direct and diffuse irradiance, on nearby terrain, that is reflected toward the point whose radiation flux we want to calculate.

Estimation of Grain Size and Detection of Absorbing Impurities

In the visible wavelengths, we should be able to estimate the extent to which the reflectance of snow has been degraded, either by absorbing impurities or by shallow depth. However, this sensitivity would be best for the blue wavelengths, where the low saturation values of the Thematic Mapper make its use for this purpose difficult. In the near-infrared wavelengths, we should be able to estimate the grain size, and thus extend the estimate of the spectral albedo throughout the solar wavelengths. Moreover, this information would help us interpret the spectral signature of snow at microwave frequencies.


Multispectral measurements in visible and near-infrared wavelengths have been used to map snow for more than two decades. Improved spectral coverage from the Landsat Thematic Mapper has allowed better estimation of snow properties and discrimination between snow cover and cloud cover. Future sensors with better spectral resolution should allow estimation of grain size and contamination by absorbing impurities, which in turn can be used to calculate spectral reflectance through the wavelengths of the solar spectrum.

Figure 1. Complex refractive index (n + i k) of ice and water

Ice data are from Warren (1984); water data are from Hale and Querry (1973), Palmer and Williams (1974), and Downing and Williams (1975). Upper: real part of refractive index (n). Lower imaginary part of refractive index (k).

Figure 2. Spectral reflectance of deep snow at illumination
angles 60 degrees and 30 degrees

The curves represent grain radii of 50 um (upper), 200 um, 500 um, and 1,000 um (lower). In the visible wavelengths (0.4 to 0.7 um) reflectance is insensitive to grain size. In the near infrared, especially from 0.9 to 1.3 um, reflectance is very sensitive to grain size. From 1.55 to 1.7 um reflectance is sensitive, but only for small sizes. The effect of illumination angle is greatest in the near-infrared.

Figure 3. Spectral reflectance of shallow snow

Figure 4. Spectral reflectance of ice and water clouds


Andersen, T., Operational snow mapping by satellites, in Hydrological Aspects of Alpine and High Mountain Areas, J. W. Glen (ed.), 149-154, International Association of Hydrological Sciences, Wallingford, UK, 1982.

Bohren, C. F. and B. R. Badcstrom, Theory of the optical properties of snow, Journal of Geophysical Rcscarch, 79, 4527-4535, 1974.

Bohren, C. F., Applicability of effective-medium theories to problems of scattering and absorption by nonhomogeneous atmospheric particles, Journal of the Atrnospheric Sciences, 43, 468-475, 1986.

Chylek, P., V. Ramaswamy and V. Srivastava, Albedo of soot-contaminated snow, Journal of Geophysical Research, 88, 10,837-10,843, 1983.

Colbeck, S. C., Grain clusters in wet snow, Journal of Colloid and Interfacc Science, 72, 371-384, 1979.

Colbeck, S. C., E. A. Anderson, V. C. Bissel, A. G. Crook, D. H. Male, C. W. Slaughter and D. R. Wiesnet, Snow accumulation, distribution, melt, and runoff, Eos, Transactions of the American Geophysical Union, 60, 46S-474, 1979.

Coibeck, S. C., Classification of seasonal snow cover crystals, Water Resources Research, 22, 59S-70S, 1986.

Davis, R. E. and J. Dozier, Stereological characterization of dry alpine snow for microwave remote sensing, Advances in Space Research, 9, 245-251, 1989.

Downing, H. and D. Williams, Optical constants of water in the infrared, Journal of Geophysical Rescarch, 80, 1656-1661, 1975.

Dozier, J. and S. I. Outcalt, An approach toward energy balance simulation over rugged terrain, Gcographical Analysis, 11, 65-85, 1979.

Dozier, J., R. E. Davis, A. T. C. Chang and K. Brown, The spectral bidirectional reflectance of snow, in Spectral Signatures of Objects in Remote Sensing, 87-92, European Space Agency, Paris, 1988.

Dozier, J., Spectral signature of alpine snow cover from the Landsat Thematic Mapper, Remote Sensing of Environment, 28, 9-22, 1989.

Dozier, J. and J. Frew, Rapid calculation of terrain pammeters for radiation modeling fmm digital elevation data, IEEE Transactions on Geoscience and Remote Sensing, 28, 963-969, 1990.

Grenfell, T. C., D. K. Perovich and J. A. Ogren, Spectral albedoes of an alpine snowpack, Cold Regions Science and Technology, 4, 121-127, 1981.

Hale, G. M. and M. R. Querry, Optical constants of water in the 200 nm to 200 ~m wavelength region, Applied Optics, 12, 55S-563, 1973.

Hyvrinen, T. and J. Lammasniemi, Infrared measurement of free-water content and grain size of snow, Opacal Engineering, 26, 342-348, 1987.

Kneizys, F. X., E. P. Shettle, L. W. Abreu, J. H. Chetwynd, G. P. Anderson, W. O. Gallery, J. E. A. Selby and S. A. Clough, Users Guide to LOWTRAN7, Report AFGL-Tech. Rep.-88-0177, Air Force Geophysics Laboratory, Bedford, MA, 1988.

Marlcham, B. L. and J. L. Barker, Thematic Mapper bandpass solar exoatmospheric irradiances, International Journal of Remote Sensing, 8, 517-523,1987.

Martinec, J. and A. Rango, Parameter values for snowmelt runoff modelling, Journal of Hydrology, 84, 197-219,1986.

Meador, W. E. and W. R. Weaver, Two-stream approximations to radiative transfer in planetary atmospheres: a unified description of existing methods and a new improvement, Journal of the Atmospheric Sciences, 37, 630-643, 1980.

Meier, M. F., Contribution of small glaciers to global sea level, Science, 226, 1418- 1421, 1984.

Nussenzveig, H. M. and W. J. Wiscombe, Efficiency factors in Mie scattering, Physical Review Letters,45, 1490-1494, 1980.

Olyphant, G. A., Insolation topoclimates and potential ablation in alpine snow accumulation basins: Front Range, Colorado, Water Resources Research, 20, 491498, 1984.

Palmer, K. and D. Williams, Optical properties of water in the near infrared, Journal of the Optical Society of America, 64, 1107-1110, 1974.

Rango, A., V. V. Salomonson and J. L. Foster, Seasonal streamflow eshmation in the Himalayan region employing meteorological satellite snow cover observadons, Water Resources Research, 14, 359-373, 1977.

Rango, A., J. Mariinec, J. Foster and D. Marics, Resoludon in operadonal remote sensing of snow cover, in Hydrological Applications of Remote Sensing and Remote Data Transmission, B. E. Goodison (ed.), 371-382, Inten~ational Association of Hydrological Sciences, Wallingford, UK,1983.

Rott, H., Remote sensing of snow, in Large Scale Effects of Seasonal Snow Cover, B. E. Goodison, R. G. Ba'Ty, and J. Dozier (ed.), 279-290, Intemational Association of Hydrological Sciences, Wallingford, UK, 1987.

Walsh, J. E., Snow cover and atmosphenc vanability, American Scientist, 72, 50-57, 1984.

Warren, S. G. and W. J. Wiscombe, A model for the spectral albedo of snow, II, Snow containing atmosphenc aerosols, Journal of the Atmospheric Sciences, 37, 2734-2745, 1980.

Warren, S. G., Optical properties of snow, Reviews of Geophysics and Space Physics, 20, 67-89, 1982.

Warren, S. G., Optical constants of ice from the ultraviolet to the microwave, Applied Optics, 23, 1206 1225, 1984.

Williams, L. D., R. G. Barry and J. T. Andrews, Application of computed global radiation for areas of high relief, Journal of Applied Meteorology, 11, 526-533, 1972.

Wiscombe, W. J. and S. G. Warren, A model for the spectral albedo of snow, I, Pure snow, Journal of the Atmospheric Sciences, 37, 2712-2733, 1980.