A0965
Title: Forecasting the atmospheric ethane burden above the Jungfraujoch with Bayesian and frequentist methods
Authors: Marina Friedrich - VU Amsterdam (Netherlands) [presenting]
Yuliya Shapovalova - Radboud University (Netherlands)
Karim Moussa - VU Amsterdam (Netherlands)
David van der Straten - VU Amsterdam (Netherlands)
Abstract: Short-lived climate forcers are broadly divided into methane and non-methane volatile organic compounds (NMVOC). They affect the climate and are often also air pollutants. Ethane is the most abundant NMVOC in the atmosphere, sharing important emission sources with methane. The main sources of ethane are anthropogenic, while methane has natural and anthropogenic sources. Understanding trends in atmospheric ethane is crucial to better constrain the anthropogenic sources of methane, in particular from the oil and gas industry. While previous studies focus on analyzing past trends, the aim is to forecast the atmospheric ethane burden above the Jungfraujoch (Switzerland), using both Bayesian and frequentist methods. Since ethane measurements can only be taken under clear-sky conditions, a substantial fraction of the data (around 75\%) is missing. The presence of missing data complicates the analysis and limits the availability of appropriate forecasting methods. Three distinct approaches for forecasting ethane time series are employed: 1) state-space modeling; 2) kernel regression which has previously been used for trend analysis in ethane time series; 3) a Gaussian process regression, which can be seen as Bayesian non-parametric regression, with interpretable compositional kernel and a spectral mixture kernel. These three approaches excel in different aspects of time series forecasting, such as flexibility, interpretability, uncertainty quantification, and handling missing data.
