Wavelet-based model for long-term forecasting of CO2 levels in atmosphere

Abstract

The CO2 levels in the atmosphere serve as an indicator for global warming. The forecasts of CO2 levels that may be expected in the foreseeable future would help in formulating credible policies as well as plans towards a sustainable future. The objective of this paper is to analyse the time series of CO2 levels observed at Mauna Loa (Hawaii) using wavelet analysis and to develop a recursive forecasting model based on wavelet decomposition. Wavelet analysis enables a decomposition of a given time series into a multi resolution series providing, in the process, an insight into the likely causative influences that operate at various scales. The main advantage of wavelet analysis is that it yields simultaneous time-frequency description of the given time series while isolating features that are localized in time as well as those occurring over a longer term time horizon. Additionally, multi resolution capability of wavelet decomposition can also reveal changes or perturbations that may be masked at a single scale. The wavelet analysis of the observed CO2 levels reveals that the trend underlying the CO2 levels is time varying and there are significant changes in the slope of the trend around 1992. In order to incorporate these changes, recursive forecasting wavelet models were developed for long term forecasting and the results reveal their superior performance over traditional models like SARIMA.