Pattern detection in multivariate climatic time series: detecting the influence of solar variability on the Earth's atmosphere
by
Charles D. Camp
California Polytechnic State University - San Luis Obispo
Coauthors: Ka-Kit Tung and Jiansong Zhou

Many climatic data records are both short and noisy: short with respect to the time scales of interest and noisy in the sense that there are many processes which interact to create the data record. Therefore, it is often difficult to extract information about the underlying processes creating the data. These issues are exacerbated when the processes of interest are not the dominant influences. If the data record consists of simultaneous measurements taken at ordered moments in time (e.g., atmospheric temperature measured monthly at different spatial locations) then multivariate time series analysis techniques can be applied. Since these techniques simultaneously analyze the full multivariate data set, they have access to more information and therefore offer a large improvement over techniques which analyze each time series independently or which analyze univariate time series constructed by spatial averaging. However, since the time series for each variable are usually highly correlated, much of this additional information is redundant. There are two fundamental and related issues: Can we use the added information of a multivariate record to better isolate the underlying processes? Can we reduce the redundancy of information by reducing the dimensionality of the data set? In other words, can we find underlying patterns which capture the fundamental behavior of the data? We will discuss some multivariate analysis techniques in the context of detecting the influence of solar variability (on decadal time scales) on the Earth's atmosphere.

Date received: April 9, 2009