Dry Snow Mapping in Finland Employing Space-Borne Passive Microwave Observations and Artificial Neural Networks

摘要

One of the major objectives of passive remote sensing applied to snow is the mapping and retrieval of geophysical parameters such as depth, mean grain size, snow water equivalent and temperature. The unknown parameters can be obtained by performing the inversion of equations relating snow parameters to the brightness temperature or employing empirical retrieval algorithms based on correlation analysis. In the first case equations involved are non-linear and the problem can be ill-posed (different sets of unknown parameters can give the same set of output brightness temperatures). In the second case, the retrieval algorithm can be valid only locally and a big amount of data may be necessary. In this study the retrieval of dry snow temperature and snow water equivalent employing 19- and 37 GHz SSM/1 measured brightness temperatures is performed by means of either supervised artificial neural networks or linear regression standard techniques. Additionally, the retrieval of snow depth is also performed for the case of a single test site, where snow depth data were available. Unsupervised neural networks are employed for the detection of dry snow and classification of different snow type categories.