CAUSALITY DETECTION FOR CLIMATE INDICES USING RESERVOIR COMPUTING METHOD

Abstract

The study discovers the causal relationships between 11 major climate indices using the Reservoir Computing method combined with the Granger approach. These indices are the Atlantic Multidecadal Oscillation, Global Mean Temperature, North Atlantic Oscillation, East Central Tropical Pacific SST, North Pacific Pattern, North Tropical Atlantic Index, Pacific Decadal Oscillation, Quasi-Biennial Oscillation, Sahel Standardized Rainfall, Southern Oscillation Index, and Tropical Southern Atlantic Index. The RC provides a powerful framework through which the causal interlinkages among these indices can be analyzed in detail and in an elaborative manner. Applying the Granger approach, we seek to unravel the causal interactions and dependencies among these climate indices and how these indices influence each other over time. The findings of the study will be displayed using a causality matrix, a binary causality matrix, and a causality network graph. The visualizations will show the presence and strength of the causal relation between indices, hence providing a whole picture of the interrelation within the climate system. This study contributes toward a general understanding of climate variability and change, thus effectively demonstrating the use of the RC method to analyze causal relationships within complex climate data sets. The results are going to improve our knowledge about the interaction of various climate indices, which will, in turn, make it possible to improve the predictability and mitigation of climate change impacts.