Snow Covered Area (SCA) Program

The program "sca" implements the decision tree algorithm for estimating snow-covered area described in "Automated Mapping of Montane Snow Cover at Subpixel Resolution from the Landsat Thematic Mapper" (by Walter Rosenthal and Jeff Dozier, Water Resources Research, 1995, in press). It is an IPW program that runs in a UNIX environment. The program is in a public ftp site. Binary code for Sun 4 workstations is in "bin", while source code is in "src".

The input is an image cube with the 6 reflective TM bands (IPW band 0 = TM Band 1, ..., IPW band 6 = TM Band 7). Command line arguments are the Earth-Sun distance in Astronomical Units, the cosine of the solar zenith angle, and the estimated path radiance (in DN) for each band.

EXAMPLE:
To classify a TM scene "input_image" acquired at 1756 GMT on May 10, 1992 for a region centered at 37:29:36 north, 119:07:09 west:

sca -u 0.845588 -d 1.01 -p 52,17,14,8,3,2 input_image > output_image

     "Sca" carries out all operations on one pixel at a time. It calculates the apparent planetary reflectance for each band and then discriminates clouds from other targets using a binary classification tree. It then subtracts the estimated path radiance for each band, and pixels not masked by the cloud classification tree are sent to a second classification tree where snow-free pixels are masked. Pixels not masked by the cloud or snow classification trees are then passed to a regression tree where fractional snow covered area is estimated.

     The output is a 2-band IPW image. Band 0 is the snow cover fraction image, with values ranging from 0 to 100 (percent snow cover). Band 1 is the cloud mask (0 = cloud, 1 = clear). Clouds are screened in the SCA image, so the mask is presented for the user's convenience.

     The algorithm was developed over the Sierra Nevada of California, but should work well over similar mid-latitude mountain ranges. There are currently some limitations that the user should be aware of.

The decision tree for masking clouds was designed to detect optically thick clouds which obscure the ground, not thinner clouds such as cirrus, through which the ground is visible.

     Mechanisms that enhance VNIR reflectance or depress SWIR reflectance may lead to the false identification of snow. Examples include water beneath a turbid atmosphere, thin clouds or fog; frozen lake surfaces and glacier ice; wet desert playas or saline soils; white river water; dark water mixed with bright shore; shallow water over a sandy bottom. Glacier ice has a spectrum similar to shaded snow and, because the algorithm was not developed to distinguish these materials, will be mapped as snow.

* This project was funded by the NASA Earth Observing System program, the U.S. Army Cold Regions Research and Engineering Laboratory, and by the Sequoia 2000 project, which is supported by the University of California and Digital Equipment Corporation.