DEVELOPMENT OF AN ALGORITHM FOR SNOW REFLECTANCE AND SNOW COVER FRACTION FROM MODIS DATA

Abstract

Varied inputs in the form of snow and weather parameters are required in scientific models pertaining to avalanche and weather forecasting. These are usually collected from various field observatories at different locations in the snowbound regions; however, the vast snow-covered areas generally occur at high altitudes, which are remote and inaccessible. As a result, the feedback of snow-cover information on snow cover and snow volume from field based scientific models is usually incomplete in various Himalayan zones. In view of the inaccessibility and harsh climate conditions of the areas of interest, remote sensing is an effective tool for both local and global retrieval of various snow cover characteristics. The major advantage of using remote sensing for mapping of snow covered areas is the availability of data at a wide range of spectral, spatial and temporal resolutions. Remote sensing techniques therefore have been used for the retrieval of a number of snow cover parameters and properties such as depth, density, velocity, volume, snow water equivalent, areal estimates etc. Among these parameters, the accurate mapping of snow cover at an operational level is extremely important it forms a crucial input for climate change models. Until recently snow cover mapping and monitoring was being carried out using satellite sensors such as NOAA AVHRR, Landsat MSS and TM, which unfortunately have become saturated for snow due its high reflectance in most of their spectral regions. Availability of data from new sensors such as MODIS on-board Terra and Aqua satellites has not only helped in overcoming the saturation problem due to their high radiometric resolutions but also facilitated the effective monitoring of large snow covered areas as a result of high temporal resolution and wide swath coverage. As it is well known, raw digital numbers (DN) derived from satellite optical systems cannot confidently be used for engineering studies, since they include effects derived from sensor calibration, as well as atmospheric and topographic interferences. Calibration values to transform DN to sensor radiance are typically provided in the image metadata, and therefore the user needs only to compensate for atmospheric and topographic effects. This is a crucial process, which must be performed in order to extract high quality information from remote sensing data. Due to the availability of remote sensing data from a variety of sensors, no one particular model may be suitable to all types of data. It is therefore important that appropriate reflectance model be developed and adopted for the data in hand and for the given application. Further, the data provided from coarse spatial resolution sensors such as MODIS are fraught with the problem of mixed pixels (i.e., pixels containing more than one class). Therefore, a fractional snow cover mapping algorithm for NOAA-AVHRR data based on an empirical `reflectance to snow-cover' relationship needs to be developed. Most of the operational snow-cover mapping algorithms involve the use of NDSI. However, the relationship between snow cover fractions and NDSI still needs to be carefully studied with the aid of field as well as reference snow cover fraction information derived from high resolution remote sensing images. Thus, in order to map snow cover accurately from other land covers, appropriate reflectance models, sub-pixel techniques and empirical relationships between the fraction snow cover and NDSI need to be developed at operational level. 111