Comparison of NDSI and LSU Methods in Estimation of Snow Cover with MODIS Sensor (Case Study: Saghez Watershed Basin

More than 30 percent of the Earth is covered by seasonal snow and about 10 percent of it by permanent glaciers. Approximately about 5 percent of global precipitation is snow. This amount to about 50 to 95 percent is in polar areas. Spatial-temporal distribution of snow is important. The estimation of snow cover area provides valuable information on snow-melting water in terms of runoff and water supply of the watershed in mountainous regions. Snow is an important geophysical factor for climate through its role in the Earth’s albedo and for hydrology. It has potentials for water storage, agriculture, hydropower generation, and flooding in local scales. Snow creates an insulating to keep plants from the cold weather in winter. Therefore, it is necessary to study the snow cover area, snow depth, and snow water equivalent. The snow cover area is affected by environmental factors, leading to different melting patterns which are important for assigning a deterministic model.
The importance of snow has been recently understood by scientists and watershed managers resulted in different snow studies. Applying remotely sensed data for such studies is cheaper faster and easier than traditional approaches. One can also study larger areas using these data which is more beneficial. Snow cover area is the most accurate factor of snow which can be estimated using remotely sensed data. Different sensors have been applied to study snow areas which have their own advantages and disadvantages. Earth Observing System (EOS)-Terra was launched with 5 mounted sensors on December 18, 1999. One of the 5 sensors in EOS is MODIS. The sensor was embedded on Aqua satellite launched on May 3, 2002. Terra is a sun-synchronous satellite, elevated at 705 Km, having polar orbit. Terra passes occur at roughly 11:00 – 12:00 AM and 10:00 – 11:00 PM local standard time each day. MODIS is the biggest sensor in EOS. Its mission is to measure temperature, ocean color, vegetation and deforestation, clouds, aerosols, and snow covers. Different ground resolution, the capability of distinguishing cloud from snow, and provide complete coverage of the Earth. Therefore, this sensor has very high potentials in snow cover studies. The sensor has a radiometric resolution of 12 bites, spectral resolution of 36 bands from 0.4 till 14.4 µ. It also has a high temporal resolution (Repeating cycle of 1 to 2 days), and moderate spatial resolution (250, 500, and 1000 m).
MODIS satellite images are used for estimating snow cover area in this study. In this research, two common techniques including Normalized Difference Snow Index (NDSI) and Linear Spectral Unmixing (LSU) were used. In order to determine the accurassy of NDSI and LSU approaches (MODIS images), the IRS images were selected since their spatial resolution is very high (24 m). The MODIS pixels were interpreted as snow using Snow map algorithm. A number of 11 similar sites on MODIS and IRS were selected to compare the results. The snow area of MODIS images (NDIS and LSU) were compared with the corresponding value on IRS images using t-student test and regression coefficients. A scatter plot of non-snow against snow was used. A regression model was established for the same purpose.
The scatter plots of snow areas produced by crossing IRS versus snow area estimated by NDSI and LSU approaches were separately studied. The regression model of each scatter plot was then calculated. The results show that both NDSI and LSU methods have high efficiency to compute snow cover area; however, the LSU method shows a little more efficiency than the NDSI method. Another comparative investigation over the NDSI and LSU methods was performed by t-student test with significant level of 5%. The t-student test indicated that the LSU method has a higher potential in estimating snow cover area in the study area than the NDSI method.
The use of remote sensing techniques, satellite images, GIS, and statistical methods for studying and monitoring ground features such as snow is very beneficial due to their lower expenses and ease of use. Among them, high temporal and spatial resolution images are preferred. Due to the importance of snow in the study area, the snow cover area was computed using MODIS and IRS satellite images to determine the best approach. The results showed that to use the methods which apply subpixels to calculate snow cover area is more appropriate. The study reveals that remote sensing techniques can provide reliable information on snow and can overcome the problems stemming from traditional approaches