An adaptive multilevel correlation analysis: a new algorithm and case study

Abstract

An adaptive multilevel correlation analysis, a kind of data-driven methodology, is proposed. The analysis is done by subdividing the time series into segments such that adjacent segments have significantly different mean values. It is shown that the proposed methodology can provide multilevel information about the correlation between two variables. An integrated coefficient with its significance testing is also proposed to summarize the correlation at each level. Using the adaptive multilevel correlation analysis methodology, the correlation between streamflow and water level is investigated for a case study, and the results indicate that real correlation might be far more complicated than the empirically constructed picture. EDITOR D. Koutsoyiannis ASSOCIATE EDITOR E. Volpi