Research On Early Warning Signals Of Abrupt Climate Change Based On Changing Probability Density Function

Historical climate records show that abrupt changes often occur in the climate system,which will have a huge impact on social economy and ecological environment.Therefore,the study of early warning signals of abrupt climate changes is very important for early warning of abrupt climate changes,which has important scientific and practical significance.Previous studies have found that in some systems with critical points,the degree of asymmetry of system state may change before the abrupt change.Therefore,some foreign researchers have proposed physical quantity which can describe asymmetry quantificationally-skewness and kurtosis,as warning signals.In this paper,the early warning performance of these two kinds of physical quantities is systematically investigated in different dynamic systems and it is found that they may fail in some models.So we use another parameter that describes the degree of asymmetry,namely the lambda coefficient of Box-Cox transform,which is expected to be a new early warning signal.Based on four simple folding bifurcation models,the lambda coefficient is found to be an early warning signal for a condition where both skewness and kurtosis fail.Comparing with skewness and kurtosis,it is found that the fluctuation of lambda coefficient is greater than that of skewness or kurtosis when all three warning signals work.Therefor,the proposed lambda coefficient in the Box-Cox transformation as an early warning indicator of abrupt climate change has certain advantages over skewness and kurtosis.In real observation data,observation error and missing data are often encountered,so we systematically studied their influence on the performance of early warning signal,respectively.The influence of sample size on early warning signal was also investigated.This paper mainly studies the influence of noise,simulating observation error,with a signal-to-noise ratio of 5dB,10 dB and 20 dB on the warning signal.It is found that noise will reduce the value of the warning signal,shorten the warning time,and the influence degree will be enhanced with the increase of noise intensity.Strong noise may even lead to complete failure of the warning signal.However,in all the experiments in our work,the missing data has almost no effect,even if the absence of measurement reaches 20% of the total sample size.With study on influence of the sample size,it is found that the sample size used to calculate the warning signal in this paper is large enough to ensure the statistical significance for skewness,kurtosis and lambda coefficient.In the first three models,the demand for sample size is almost thesame,while in the fourth model,the demand for sample size of lambda coefficient is smaller than that of skewness and kurtosis.In the application of the observation data of the Forbush Decrease(FD),the three early warning signals began to show a continuous change trend about four days in advance,that is to say,each of them play an early warning role.