DISCOVERY OF CLIMATE PATTERNS AND THEIR SPATIO-TEMPORAL EVOLUTION

Abstract

With the availability of a large amount of data available from satellites, sensors, physical models etc., climate science has emerged as an important field of data science. Climate pattern discovery focuses on discovering interesting associations among climate variables with applications ranging from predicting the future climate trends to understanding the climate phenomenon. One of the grand challenges in Climate pattern discovery is to characterize the spatio-temporal evolution of the climate. However, existing work done in this domain has failed to address the issue owing to the complexity of spatio-temporal climate dataset. In this thesis, we address the mentioned problem of tracking the evolution of climate. We introduce SteC patterns as an effective solution for tracking the climate evolution. STeC patterns explicitly account for the dependency between climate variables, along with the temporal and spatial profiling. We suggest a technique based on two level Self-Organizing Maps for data driven discovery of the STeC patterns from climate data. The two level Self-Organizing Map is prepared using real world earth climate data. The SteC patterns are then generated by traversing the two level structure from top to bottom. The generated SteC patterns have been tested against various ground truths for validation. The outcome of experimental validation suggests that the STeC patterns are effective in capturing the eftect of various climate phenomenon such as teleconnection patterns, climate changes etc. We have also discussed the use of our technique for identifiing growing and shrinking patterns, stable regions etc.