Integration of Special Sensor Microwave/Imager (SSM/I) and in Situ Data for Snow Studies from Space

• Main Author: Sun, Changyi
• Format: Others
• Published: DigitalCommons@USU 1996
• Subjects: special sensor microwave; imager; ssm/i; situ data; space; Forest Sciences
• Online Access: https://digitalcommons.usu.edu/etd/7297; https://digitalcommons.usu.edu/cgi/viewcontent.cgi?article=8394&context=etd

The Special Sensor Microwave/Imager (SSM/I) radiometer is a useful tool for monitoring snow conditions and estimating snow water equivalent and wetness because it is sensitive to the changes in the physical and dielectric properties of snow. Development and improvement of SSM/I snow-related algorithms is hampered generally by the lack of quantitative snow wetness data and the restriction of a fixed uniform footprint. Currently, there is a need for snow classification algorithms for terrain where forests overlie snow cover. A field experiment was conducted to examine the relationship between snow wetness and meteorological variables. Based on the relationship, snow wetness was estimated concurrently with SSM/I local crossing time at selected footprints to develop an SSM/I snow wetness algorithm. For the improvement of existing algorithms, SSM/I observations were linked with concurrent ground-based snow data over a study area containing both sparse- and medium-vegetated regions. Unsupervised cluster analysis was applied to separate SSM/I brightness temperature (Tb) data into groups. Six typical SSM/I Tb signatures, based on cluster means of desired snow classes, were identified. An artificial neural network (ANN) classifier was designed to learn the typical Tb patterns Ill for land-surface snow cover classification. An ANN approximator was trained with the relations between inputs of SSM/I Tb observations and outputs of ground-based snow water equivalent and wetness. Results indicated that snow wetness estimated from concurrent air temperature could provide the ground-based data needed for the development of SSM/I algorithms. The use of cluster means might be sufficient in ANN supervised learning for snow classification, and the ANN has the potential to be trained for retrieving different snow parameters simultaneously from SSM/I data. It is concluded that the ANN approach may overcome the drawbacks and limitations of the existing SSM/I algorithms for land-surface snow classification and parameter estimation over varied terrain. This study demonstrated a nonlinear retrieval method towards making the inferences of snow conditions and parameters from SSM/I data over varied terrain operational.