Applications Of Time Series Network Methods To Climate Drought Events Analysis

Entering the big data era of economic,technological and cultural globalization,the application of complex network and nonlinear time series research gets more and more attention,such as the epidemic spread,the brain functional networks,extreme climate events as well as the financial stock volatility.Time series analysis methods can provide us with innovative thinking and exploration direction.In the present,traditional nonlinear time series methods are experiencing rapidly development due to the influence of complex network research.However,wide applications of time series of network methods are greatly restricted because of the influence of series length and the noise effects in practical researches.Taking time series of climate drought extreme events as examples,we focus on discussing the common problems encountered by complex network methods in real time series analysis and providing solutions.This thesis focuses on the ongoing drought events in northeast Brazil(NEB)in recent years because this region has been experiencing a persistent drought since 2010,resulting in significant social impacts and economic losses.The physical mechanisms of extreme drought events in the NEB are due to varying interacting timescales from the surrounding tropical oceans.We characterize the influence from the ocean to the precipitation in the NEB by ordinal partition transition network approaches.In particular,we will focus on the Tropical South Atlantic and ENSO regions in the Pacific.This approach converts the event time series into order pattern symbol sequence in order to construct ordinal partition transition network.In addition,we propose to calculate complexity measures of inhomogeneity between symbol sequences by the divergence of Kullback-Leibler(KLD),capturing the coupling time delay information between two time series.In addition,this paper proposes a hypothesis test of statistical significance based on KLD calculation,which is well verified.Finally,we quantitatively analyze the intra-seasonal time scale dynamic changes of precipitation in NEB by using time-delay information and drought periodlength(DPL)that are extracted by KLD measures.Nonlinear time series network method proposed in this paper provides new insights for studying the interaction between two event time series,and the statistical significance of the detected coupling direction has been assessed by the relevant statistical hypothesis testing,which therefore are important for the follow-up research on the coupling relationship between multiple systems.