AbstractA comprehensive change-point analysis of annual radiosonde temperature measurements collected at the surface, troposphere, tropopause, and lower-stratosphere levels at both the South and North Polar zones has been done. The data from each zone are modeled as a multivariate Gaussian series with a possible change point in both the mean vector as well as the covariance matrix. Prior to carrying out an analysis of the data, a methodology for computing the large sample distribution of the maximum likelihood estimator of the change point is first developed. The Bayesian approach for change-point estimation under conjugate priors is also developed. A simulation study is carried out to compare the maximum likelihood estimator and various Bayesian estimates. Then, a comprehensive change-point analysis under a multivariate framework is carried out on the temperature data for the period 1958–2008. Change detection is based on the likelihood ratio procedure, and change-point estimation is based on the maximum likelihood principle and other Bayesian procedures. The analysis showed strong evidence of change in the correlation between tropopause and lower-stratosphere layers at the South Polar zone subsequent to 1981. The analysis also showed evidence of a cooling effect at the tropopause and lower-stratosphere layers, as well as a warming effect at the surface and troposphere layers at both the South and North Polar zones.
Step change point estimation in the multivariate-attribute process variability using artificial neural networks and maximum likelihood estimation
