Interannual Temperature Events and Shifts in Global Temperature: A “Multiwavelet” Correlation Approach

For the purpose of climate signal detection, we introduce a method for identifying significant episodes of large-scale oscillatory variability. The method is based on a multivariate wavelet algorithm that identifies coherent patterns of variation simultaneously within particular ranges of time and periodicity (or frequency) that may vary regionally in the timing and amplitude of the particular temperature oscillation. By using this methodology, an analysis is performed of the instrumental record of global temperatures spanning the past 140 years. The duration of an ''episode'' is chosen to correspond to 3-5 cycles at a specified oscillation period, which is useful for detecting signals associated with the global El Nin o/Southern Oscillation (ENSO) phenomenon. To confirm the robustness of signals detected in the earliest, sparse data (only 111 5 longitude by 5 latitude grid points are available back to 1854), we performed multiple analyses overlapping in time, using increasingly dense subsets of the full (1570 grid point) temperature data.