Application of snow distribution models within the Laguna Negra basin, Chile
PublisherThe University of Arizona.
RightsCopyright © is held by the author. Digital access to this material is made possible by the University Libraries, University of Arizona. Further transmission, reproduction or presentation (such as public display or performance) of protected items is prohibited except with permission of the author.
AbstractSpectral linear unmixing and binary regression trees were used to estimate the distribution of snow within the Laguna Negra basin in Chile. Spectral linear unmixing was performed for multi-band Landsat 5 images for the determination of sub-pixel snow fractions. We were interested in determining the number of bands needed for an adequate distribution of SCA. Results showed that for winter scenes (scenes with greater than 90% snow cover and portions of the basin covered by shadows) linear spectral unmixing can be used to model SCA using at least four bands with a rock, a snow and a shaded snow endmember, but that five bands, using two rock endmembers, a snow and a shaded rock endmember, are needed for the fall scenes (scenes with less than 10% snow cover and portions of the basin covered by shadows). The spring scenes (scenes with 50 percent and higher snow cover and no shadows) showed plausible results with three bands, but the need for a second rock endmember in the fall scenes suggest 4 bands may give a more accurate result. A binary regression tree model was used to determine distributed SWE at peak accumulation in the Echaurren basin, a sub basin of Laguna Negra. Regression trees grown from field snow survey data did an excellent job at explaining the variation of SWE in two of the three surveys examined when resubstitution was used to evaluate the model, but did a poor job in all cases when cross validation was used. However, cross validation may over estimate the errors associated with the model. Basin-wide SWE maps resulting from the application of the regression trees formed plausible structures. Normalized snow distribution was sufficiently different between years such that a "typical" SWE map could not be developed. Nonetheless, there were identifiable patterns that did occur in the SWE distributions from different years that gave insight into the factors affecting SWE in the basin. Such factors include a strong dependance on radiation in the lower portion of Echaurren for two of the years, and the presence of heavy SWE regions near cliffs. Insights such as these provided useful information on how the type of data and method of collection might be improved. The large SWE values near cliffs, for instance, suggest that use of an avalanche map might improve the modeled SWE distribution. The dependance of SWE on radiation in the lower basin suggest the SWE data should be obtained over the entire range of radiation values in the lower basin.
Degree ProgramGraduate College
Hydrology and Water Resources