Forecasting atmospheric ethane : application to the Jungfraujoch Measurement Station

Understanding the developments of atmospheric ethane is essential for better identifying the anthropogenic sources of methane, a major greenhouse gas with high global warming potential. While previous studies have focused on analyzing past trends in ethane and modeling the inter-annual variability, this paper aims at forecasting the atmospheric ethane burden above the Jungfraujoch (Switzerland). Since measurements can only be taken under clear sky conditions, a substantial fraction of the data (around 76%) is missing. The presence of missing data together with a strong seasonal component complicates the analysis and limits the availability of appropriate forecasting methods. In this paper, we propose five distinct approaches which we compare to a simple benchmark - a deterministic trending seasonal model - which is one of the most commonly used models in the ethane literature. We find that a structural time series model performs best for one-day ahead forecasts, while damped exponential smoothing and Gaussian process regression provide the best results for longer horizons. Additionally, we observe that forecasts are mostly driven by the seasonal component. This emphasizes the importance of selecting methods capable of capturing the seasonal variation in ethane measurements.