GIS Ag Maps.org

Landsat NDSI Snow Map Development Details: example from Livingston, MT area

Related pages:  Snow Mapping Example; NDSI    Snowiest pixels determination

Landsat imagery can be used to map snow extent (Dozier, 1989) based on an NDSI value of 0.40. Also, blue surface reflectance should be greater than about 0.15 to 0.20 (distinguishes snow from different surfaces in shadowed areas) and band 5 surface reflectance should be less than 0.20 to 0.25 (distinguishes snow from clouds).

For Landsat 7 & 8 purposes (and scale applied) here, we have found that it is not necesarry to exclude the blue and SWIR values described in the above paragraph for clear skies, but feel it is necessary to apply a SWIR constraint for cloudy skies. We have found that excluding pixels with NIR surface reflectance less than about 0.04 to 0.05 is an effectice method to exclude water, while including snow and do apply this method. IMPORTANT: A small adjustment should be made to Landsat 8 NDSI classification values due to its slightly smaller SWIR average band wavelength, which results in a slightly higher NDSI value for Landsat 8 than Landsat 7 for the same snow conditions. Whereby, NDSI 0.40 represents snow for Landsat 7 (Dozier [1989] agreed with this value for Landsat 5 which has the same green and SWIR [band 5] wavelength as Landsat 7), a value of 0.45 is used for Landsat 8. 

The following example shows how Landsat can be used for snow mapping near Livingston, MT. Landsat 5 imagery from 11/11/2011 is used. The normalized difference snow index (NDSI) is applied for snow mapping and is written as: (green - short wave infrared) / (green + short wave infrared). For Landsat this is: (band 2 - band 5) / (band 2 + band 5). Negi et al. (2009) found that calculating NDSI with planetary reflectance (as opposed to digital numbers) resulted in more accurate snow mapping; the atmospheric correction method to retrieve reflectance used here is described in the link on the top menu. (Dozier calculated values as apparent reflectance [atmospheric scatter was not deducted].) Dozier (1989) studied snow with Landsat in the Sierra Nevada Mountains in California while Negi et al. (2009) applied AWiFS in a basin of the Indian Himalaya. Both found that a NDSI ≥ 0.4 could be used as a criteria for snow mapping. Importantly, Negi et al. found that "NDSI values remain constant with the variations in slope and aspect and thus NDSI can take care of topography effects". NDSI can delineate and map snow in mountain shadows (Kulkarni et,. 2002).

 Livingston, MT area  

Landsat 5 TM band 2 (green) 11/11/2011; snow has a high reflectance is all visible bands (that is why is appears white.) (For all images that follow, lighter shades of gray are higher values, symbolized from ±2 s from the mean.)

Landsat 5 TM band 5 (short wave infrared [SWIR]) 11/11/2011; snow has a low reflectance in the SWIR band. 

Criteria is applied based on Dozier (1989) that band 5 maximum reflectance threshold should be 0.2 to 0.25. The green area below covers pixels with reflectance < 0.2.

Criteria is applied that NDSI should be ≥ 0.4 which is supported by Dozier (1989) and Negi et al. (2009). The green area below covers pixels with an NDSI value ≥ 0.4.

The intersection (common surface) where band 5 is < 0.2 and NDSI is ≥ 0.4 produces the snow map below; lighter areas represent higher NDSI values and more snow.

 References

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

Kulkarni A. V., Srinivasulu J., Manjul S. S. and P. Mathur. 2002. Field based spectral reflectance to develop NDSI method for the snow cover; J. Indian Soc. Remote Sens. 30(1&2): pp. 73–80. Cited in: Negi et al (2009) shown below.

Negi, H.S., Kulkarni, A.V., and B.S. Semwab. 2009. Estimation of snow cover distribution in Beas basin, Indian Himalaya using satellite data and ground measurements. J. Earth Syst. Sci. 118, No. 5: pp. 525–538.