Visualizing gridded time series data with self organizing maps: An application to multi-year snow dynamics in the Northern Hemisphere

Gridded time-series data are increasingly available for climatology research. Microwave imagery of snow water equivalent (SWE) has been accumulated at daily basis for over two decades, but complex spatial-temporal patterns in SWE dataset pose great challenges for exploration and interpretation. This paper introduces the use of several perspectives from a tri-space conceptualization of a time series of SWE grids combined with dimensionality reduction via the self-organizing map (SOM) method. While SOM has been predominantly viewed as a clustering mechanism within climatology research, we expand the visual-analytic potential of SOM for climate research with a series of conceptual, computational, and visual transformations. Specifically, we apply a medium-resolution SOM to an SWE dataset covering the Northern Hemisphere over a 20-year period, with high temporal resolution. Through clustering and visualization a number of distinct SWE patterns are identified, including mountainous, coastal, and continental regions. A subset of cells from six areas are selected for transition analysis, including mountainous (Sierra Nevada, Western Himalaya, Eastern Himalaya) and continental (central Siberia, western Russia and Midwest United States) regions. By combining with trajectory analysis, this SOM documents notable transitions in seasonal SWE accumulation and melt patterns in mountain ranges, suggesting that SWE in some geographic locations alternates between different discrete annual patterns. In the Sierra Nevada, transitions to classes with high SWE are shown to be related to the Southern Oscillation Index, demonstrating that the annual patterns and transitions in SWE regime identified by the SOM correspond to synoptic climate conditions.