Improving a plot-scale methane emission model and its performance at a northeastern Siberian tundra site
Abstract. In order to better address the feedbacks between climate and wetland methane (CH4) emissions, we tested several mechanistic improvements to the wetland CH4 emission model Peatland-VU with a longer Arctic data set than any other model: (1) inclusion of an improved hydrological module, (2) incorporation of a gross primary productivity (GPP) module, and (3) a more realistic soil-freezing scheme.
A long time series of field measurements (2003–2010) from a tundra site in northeastern Siberia is used to validate the model, and the generalized likelihood uncertainty estimation (GLUE) methodology is used to test the sensitivity of model parameters.
Peatland-VU is able to capture both the annual magnitude and seasonal variations of the CH4 flux, water table position, and soil thermal properties. However, detailed daily variations are difficult to evaluate due to data limitation. Improvements due to the inclusion of a GPP module are less than anticipated, although this component is likely to become more important at larger spatial scales because the module can accommodate the variations in vegetation traits better than at plot scale.
Sensitivity experiments suggest that the methane production rate factor, the methane plant oxidation parameter, the reference temperature for temperature-dependent decomposition, and the methane plant transport rate factor are the most important parameters affecting the data fit, regardless of vegetation type. Both wet and dry vegetation cover are sensitive to the minimum water table level; the former is also sensitive to the runoff threshold and open water correction factor, and the latter to the subsurface water evaporation and evapotranspiration correction factors.
These results shed light on model parameterization and future improvement of CH4 modelling. However, high spatial variability of CH4 emissions within similar vegetation/soil units and data quality prove to impose severe limits on model testing and improvement.