Detection Of Regime Shift And Classification Of Glacial Stadials

Since the Industrial Revolution,human production activities have released large amounts of carbon dioxide(CO₂)into the Earth's system at a rapid rate,forcing the planet's temperature to rise and the climate system to evolve towards a warmer state.Although historical periods have not seen exactly the same climate conditions as today,past climates provide an important basis for understanding global warming,so looking at ancient climate change patterns is an important way of exploring future climate trends.This paper focuses on the geological carrier record data,mainly on the filtration of multi-millennial scale variations in the carrier record and denoising of sub-millennial scale perturbations,and on the identification of millennial scale events in the ice cores record through the detection of regime shift in the time series,finally on the classification of glacial stadials.The data are oxygen 18 isotope records from the North Greenland Ice Core Project(NGRIP),temperature records from the European Project for Ice Coring in Antarctica(EPICA)in the Dronning Maud Land(EDML)and temperature records from the Dome C(EDC).Firstly,this paper performs a comparative analysis of five commonly trend extraction methods.Then three detection methods of regime shift are used to date climate events in the ice core record from NGRIP,the ice core record from EDML,and the ice core record from EDC respectively.Finally,the different glacial stadials periods are statistically classified.The results are as follows:(1)The moving average method,Savitzky-Golay filter method,empirical modal decomposition method,ensemble empirical modal decomposition method,butterworth filter,finite impulse filter designed by hanning window and wavelet decomposition and reconstruction methods are compared for the filtering of multi-millennial scale variations and denoising of low-millennial scale perturbations.Moving average method and Savitzky-Golay filter method were found to extract multi-millennial scale climate variation with burr,in which many low-target-scale variation information was mixed;empirical modal decomposition and ensemble empirical modal decomposition could not adequately fit the trend at both ends of the sequence,which was lower than the remaining four methods;finite impulse filter designed by hanning window was susceptible to extreme values,and digital filters approximate the series by steady,different-scale sine or cosine waves,are not suitable for non-stationary periodically varying climate series;the wavelet decomposition and reconstruction method is similar to a band-pass filter that decomposes the series into subseries with different frequency bands through iteration,and it can clearly capture the lowest sea level of Last Glacial Maximum when extracts multi-millennial scale climate variation information from NGRIP ice core record.Therefore,this paper uses wavelet decomposition and reconstruction method to extract multi-millennial scale climate variation as well as to remove low-millennial scale perturbations.(2)The three regime shift detection methods were compared and analysed to determine the start and end times of each climate events.According to the mechanism of bipolar seesaw,when the DO event at Greenland in a cold phase,the Atlantic meridional overturning circulation will weaken or even shut down,and therefore heat will accumulate in the South Atlantic region,leading to warming in Antarctic region with an AIM event.From the characteristics of the events,it can be seen that DO events are mean shift and AIM events are trend shift,where the heuristic segmentation algorithm does hypothesis testing on the mean in the subsequence and can only detect mean shift,and the Mann-Kendall method does hypothesis testing on the order statistics of positive and inverse sequences and can only detect trend shift.In this paper,the Ramp Fit method is used to detect mutation points in sliced ice core data separately,and to determine the start and end of each event based on the characteristics of the DO and AIM events,with the DO21 event,based on the duration of the event and the response of the Antarctic region to it,being considered a HS event.(3)An regression through the orign model was used to fit the Antarctic response to a normal DO event and the Antarctic response to a DO event with an HS event separately,and found to be significantly different at a significance level of 0.2.In summary,the results of this paper show that wavelet decomposition and reconstruction methods can decompose sequences into subseries with different frequency bands,which can be used to extract climate variations at multi-millennial scales and to remove perturbations at low millennial scales.Ramp Fit method fits a piecewise function to the subseries,which can represent both mean and trend regime shift,and could meet the requirements of detecting multiple regime shift patterns compounded in the glacial climate.The Bipolar seesaw model illustrates that the Antarctic region responds to climate events in Greenland.It is assumed that when there is no DO event in the Northern Hemisphere,i.e.no sustained cold phase,the AMOC will not weaken or close and the Southern Hemisphere will not accumulate heat for warming.