[M-GI27] Data-driven approaches for weather and hydrological predictions

[MGI27-04] Sequential analysis of tipping in high-dimensional complex systems with partially known dynamics

*Tomomasa Hirose¹, Yohei Sawada¹ (1.The University of Tokyo)

Keywords:Earth system, climate change, tipping, chaos

Recently, the term “climate tipping” has been drawing many scientists’ attention. This term refers to abrupt, often irreversible changes in the Earth system. Although many theoretical works have been done to analyze and to predict this phenomena, various unsolved problems still exist towards analyzing real climatological data. Among them, we focused on two primary challenges: imperfect knowledge of the system and high-dimensionality. To tackle these problems, we proposed a tipping analysis framework called DA-HASC (Data Assimilation-High dimensional Attractor’s Structural Complexity). First, from limited observation data and partial knowledge on the system dynamics, we reconstruct a high-dimensional state by data assimilation technique. Second, we split reconstructed time-series data into windows and quantify each local attractor’s complexity to capture underlying change in the high-dimensional system’s dynamics. In this second step, we adopted manifold learning technique to preserve high-dimensional structural information. The information is provided as graph representation, which is later measured by Von Neumann entropy. The framework was evaluated by both synthetic and real-world data and showed promising performances to detect tipping of high-dimensional partially known dynamics.