A Bayesian approach to gapfilling fluxes from heterogenous Dutch peatlands measured by eddy covariance

Peatlands are commonly a mosaic of vegetation types, ditches, and open water. Eddy covariance provides a useful way to measure fluxes at the ecosystem scale, however comparing fluxes and annual balances from different source areas in heterogenous field sites can be challenging. Separating fluxes based on wind direction is a commonly used approach to study different source areas, but this method may exclude a substantial amounts of data points that contain some information about flux behaviour, such as when the flux source area covers multiple land cover classes, and as a result the gapfilled timeseries contains greater uncertainty. In this presentation, we present a Bayesian approach that utilises the flux footprint to gapfill CO₂ and CH₄ fluxes. A flux footprint model was used to predict the flux source area and the relative contribution for each land cover class is calculated for each timestep. The net ecosystem exchange (NEE) of a scalar is then assumed to be the linear combination of the different land cover classes weighted by the contribution within the flux footprint. A Bayesian framework was used to estimate model parameters for gapfilling each land cover class for the NEE of CO₂ and CH₄, where the non-linear model approaches were used in both cases. For CO₂, the framework estimated the parameters for the gross primary production (GPP) light response curve and R_eco via the Lloyd-Taylor respiration function, where the NEEof CO₂ was then calculated as the sum of GPP and R_eco. For CH₄, a non-linear temperature dependence model was used.  We show results from multiple eddy covariance towers on peatlands in the Netherlands, including a mixed paludiculture pasture site, natural vegetation sites, and a pasture site with subsurface drainage. We demonstrate that this approach is useful for constraining flux behaviour and obtaining annual balances for each land cover class within the flux footprint.