Adaptive time series analysis of Mauna Loa CO2 data: tvf-EMD based detrend and extraction of seasonal variability

TL;DR

This paper applies a novel adaptive time series analysis method, tvf-EMD, to Mauna Loa CO2 data, successfully extracting trends and seasonal cycles while analyzing residuals and outliers in relation to solar activity.

Contribution

The study develops a Python implementation of tvf-EMD and demonstrates its effectiveness in separating trend and seasonal components in noisy CO2 data without mode mixing.

Findings

Successfully extracted trend, seasonal, and residual components from CO2 data.

Residuals follow a normal distribution and show increased oscillation during solar cycle minima.

Outliers are more frequent towards the end of the dataset, correlating with solar activity.

Abstract

Adaptive time series analysis has been applied to investigate variability of CO2 concentration data, sampled weekly at Mauna Loa monitoring station. Due to its ability to mitigate mode mixing, the recent time varying filter Empirical Mode Decomposition (tvf-EMD) methodology is employed to extract local narrowband oscillatory modes. In order to perform data analysis, we developed a Python implementation of the tvf-EMD algorithm, referred to as pytvfemd. The algorithm allowed to extract the trend and both the six month and the one year periodicities, without mode mixing, even though the analysed data are noisy. Furthermore, subtracting such modes the residuals are obtained, which are found to be described by a normal distribution. Outliers occurrence was also investigated and it is found that they occur in higher number toward the end of the dataset, corresponding to solar cycles characterised by smaller sunspot numbers. A more pronounced oscillation of the residuals is also observed in this regard, in relation to the solar cycles activity too.