Master's Project
The environment is changing due to anthropogenic carbon emissions, and so is the carbon cycle regulating the exchange of CO2 (i.e. fluxes) between the Earth’s surface and the atmosphere. Measuring these changes is difficult, as it would require enormously dense observation networks to capture the strongly heterogeneous underlying flux-landscape. Through a combination of carbon exchange (CE) models and data assimilation (DA), the CarbonTracker data assimilation shell (CTDAS) generates a flux-landscape estimate which optimally matches the available observations. The current implementation of this DA approach is static; flux-landscape estimates produced in the past are not used for estimating new flux-landscapes. However, preliminary research has shown that seasonal, currently unused, patterns are present within the estimates of the DA approach. We propose three different methods for utilizing these patterns: a simple monthly mean model, a seasonal autoregressive integrated moving average (SARIMA) model, and a seasonal autoregressive integrated moving average with exogenous factors (SARIMAX) model. Preliminary results strongly indicate that the monthly mean model provides a substantial improvement over the current DA implementation once incorporated within CTDAS. In contrast, the SARIMA and SARIMAX models struggle to capture the non-stationary seasonal patterns.
The full thesis can be found here, and the source code is found on the GitHub repository.