Identifying strong signals between low‐frequency climate oscillations and annual precipitation using correlation analysis

Abstract

Long‐term changes in precipitation in California and North and South Carolina are correlated to low‐frequency oscillations of several hydroclimate indices (HCIs) through correlation analysis that utilizes longer sliding window sizes compared to previous studies to reduce higher‐frequency noise in each time series. HCIs that are considered include the El Niño/Southern Oscillation (ENSO), the Madden–Julian Oscillation (MJO), the North Atlantic Oscillation, the Pacific‐Decadal Oscillation, among others. Multi‐year accumulations of precipitation at several point locations were correlated to these HCIs temporally averaged over the same period. The sliding window size, lag time, and beginning month were varied to optimize the correlation for each site and HCI; a 60‐month window size and 12‐month lag time were found to result in the highest correlation. Correlation strength was characterized by the Pearson's r statistic, while correlation significance was estimated through a permutation experiment that employs a bootstrapping technique, resulting in a p‐value between 0 and 1. Using a 60‐month sliding window size and 12‐month lag time, it was found that the MJO exhibited the strongest and the most significant correlation with accumulated precipitation throughout California, whereas similar correlation was found with ENSO throughout the Carolinas; correlation strength exceeded a Pearson's r of .80, while correlation significance was p < .05 at several sites. Optimal beginning months ranged from December to March for a majority of sites. This study underscores the potential of low‐frequency climate oscillations that manifest themselves in the long‐range dependence of precipitation on tropical disturbances.

Characterization of links between hydro‐climate indices and long‐term precipitation in Brazil using correlation analysis

“…When all stations are considered, the average correlation strength between monthly precipitation and the optimal HCI shows a peak at a sliding window size of 60 months, with a lag time of 12 months (see Figure 2). These results are consistent with those shown by Giovannettone and Zhang (2019), who found similar values in California and North and South Carolina within the United States.…”

“…In this context, we hypothesize that the mechanisms by which these lower‐frequency oscillations affect precipitation within Brazil are similar to those associated with the higher‐frequency 30–90 day cycle. Consequently, the lower‐frequency oscillations favour an enhancement or a suppression of these mechanisms and result in higher 30–90 day highs and lower 30–90 day lows (Giovannettone & Zhang, 2019).…”

“…The methodology and tools used in the current study to assess correlation strength and significance between rainfall at each INMET site and a host of HCIs are described in detail in Giovannettone and Zhang (2019). In summary, the method requires estimating the correlation strength of rainfall with each HCI using a range of sliding window sizes (1–90 months), lag times (1–60 months), and beginning months (January–December) that define the precipitation and HCI sliding windows.…”

“…Using the combination of the aforementioned parameters, a HCI exhibiting the highest correlation (thereafter referred to as dominant) was identified at each site. The HydroMetriks/Frequency Intensity Tool (Hydro‐FIT; Giovannettone, 2017b; Giovannettone & Zhang, 2019) was used to facilitate this analysis over such a large dataset. Because the above correlation analysis results in reduced record length and potentially enhanced serial correlation (Sakumura, Bettadpur, Save, & McCullough, 2016; Upham, Cambouropoulos, Tsougras, Mavromatis, & Pastiades, 2012), a bootstrapping technique developed by Efron (1992) and described in Giovannettone and Zhang (2019) was used to test the significance of each correlation for all INMET sites (211 sites).…”

“…The HydroMetriks/Frequency Intensity Tool (Hydro‐FIT; Giovannettone, 2017b; Giovannettone & Zhang, 2019) was used to facilitate this analysis over such a large dataset. Because the above correlation analysis results in reduced record length and potentially enhanced serial correlation (Sakumura, Bettadpur, Save, & McCullough, 2016; Upham, Cambouropoulos, Tsougras, Mavromatis, & Pastiades, 2012), a bootstrapping technique developed by Efron (1992) and described in Giovannettone and Zhang (2019) was used to test the significance of each correlation for all INMET sites (211 sites). The final result is represented by a p ‐value or probability that the correlation may be a random artefact of the statistical analysis that lacks a causal basis (Shao & Tu, 2012).…”