Spatio-Temporal Reconstruction of the Global CO2-Fluxes Using Gaussian Markov Random Fields

• Activity Number: 264
• Type: Contributed
• Date/Time: Monday, August 10, 2015 : 2:00 PM to 3:50 PM
• Sponsor: Section on Statistics and the Environment
• Author(s): Unn Dahlén* and Johan Lindström
• Companies: Lund University and Lund University
• Keywords: Gaussian Markov Random Fields ; Carbon dioxide Flux ; Inverse modelling ; Global data ; Non-stationary Fields

Abstract

Reconstruction of local or regional fluxes of Carbon dioxide is important in assessing the contribution of different areas to the global carbon balance. These reconstructions are based on measurement of atmospheric concentrations and transport (or sensitivity) matrices, which link local fluxes to measurements. A common practise (e.g. R\"{o}denbeck et al, 2003) is to model the fluxes as a stationary latent Gaussian field with exponential covariance function and Gaussian observations. Commonly, the estimate of the posterior uncertainty is based on solely prior information, which might be seen as an inefficient use of observations. We model the latent field using a non-stationary Gaussian Markov Random Field defined as the solution to a stochastic partial differential equation on the globe. This formulation allows for a class of covariance functions that naturally account for the spherical data while the non-stationarity makes it possible to account for differences in correlation strength due to e.g. latitude and land/ocean interactions.