Carbon Dioxide Time Series Analysis: A New Methodological Approach for Event Screening Categorisation

Abstract

A new method for time series analysis allowing the description of background evolution and outliers was developed. This approach was tested on CO2 weekly measurements made at Mauna Loa, Hawaii, and Lampedusa, Italy, for a period (1992–2014) longer than the 11-year solar cycle. After the time series was detrended, the Generalised Lomb-Scargle periodogram was computed for frequency domain analysis. All frequencies corresponding to values in the spectrum higher than a threshold were filtered out, and the time series was reduced to residuals. Residuals were analysed principally focusing on persistency, inspected via detrended fluctuation analysis. The analysis allows to highlight similarities and differences between the two stations. Annual and semi-annual periods are present in both the time series, with significantly larger amplitudes at Lampedusa than at Mauna Loa, where however they explain about 83% of the variability, compared to about 62% at Lampedusa. Remarkably, a different Hurst exponent was found, with a value corresponding to pink noise for Mauna Loa, and a smaller value (about 0.80) for Lampedusa. This is attributed to the different characteristics of the two stations.