Visualization of huge climate data with high-speed spherical self-organizing map

We propose the use of a high-speed spherical selforganizing map (HSS-SOM) to visualize climate variability as a complementary alternative to empirical orthogonal function (EOF) analysis. EOF analysis, which is the same as principal component analysis, is often used in the fields of meteorology and climatology to extract leading climate variability patterns, its production of linear mapping with only a low contribution rate may preclude producing any meaningful results. Due to computational limitations, however, conventional self-organizing maps are difficult to apply to huge climate datasets. The development of HSSSOMs with dynamically growing neurons has helped reduce computational time. After demonstrating validity of our HSS-SOM using observational climate data and HSS-SOM effectiveness as a complementary alternative to the EOF, we extract dominant atmospheric circulation patterns from huge amounts of climate data in the general circulation model, in which both present climatology and future climate are simulated. These patterns correspond to those obtained in previous studies, indicating the HSS-SOM's usefulness in climate research.