The four-fold cyclic periodicity of the Vostok ice core data in estimating temperature and gas concentration allows for a detailed time-frequency (harmonic) as well as energy analysis in the context of machine learning data set training and testing and building short- and long-term global climate variables predictions. Assuming causal time regularity, the more of these cycles that are employed in training the more the error in prediction for the next cycle should be reduced. In this paper we conduct harmonic and correlation analysis for temperature and carbon dioxide concentration and Vostok data energy analysis. This can be considered as a data conditioning step for a Machine Learning approach which follows. Time correlation analysis allows us to more accurately estimate the time lags in each cycle already noted between temperature change and carbon dioxide content of the global atmosphere. We estimate these lags to lie between 1000-3000 years, longer than previously concluded. The analysis reported here will be used in our future work where we will conduct Machine Learning style training sessions on various cycles and combinations of cycles and then use them to run tests for prediction sessions of existing cycles or to make short (years) and long term (100s of years) global climate predictions. We also anticipate employing Kalman filter like time models for prediction purposes. Such predictions could ultimately allow detection of climate related anomalies and trends.