Optimization of a classification scheme for Arctic wetlands to support carbon and energy flux upscaling

Wetlands play an important role in the carbon balance of the Arctic. All wetland types are characterized by an individual combination of hydrological conditions, soils, vegetation cover, etc. These characteristics influence their feedback with current and future climate conditions, and therefore also their greenhouse gas exchange processes. While most climate models distinguish only one or two types of wetlands, biogeographical approaches define at least ten types of wetlands in the Arctic.

Improving the representation of wetland ecology in carbon upscaling studies in the Arctic requires finding the balance between the diversity of wetlands, including their variability in responses to climate forcing, and the information that is commonly available to represent them in modelling frameworks. On the one hand, a larger number of classes allows a more precise description of the conditions and characteristics of the fluxes within each class. On the other hand, more classes also mean less information per class, and thus more gaps that need to be interpolated.

To support the development of a refined classification scheme, we first built a database on Arctic wetland characteristics, including measured carbon pools and fluxes, based on the available information from published studies. Our database covers the period 1988 – 2019, with observations for all seasons available. Most data was taken from flux chamber studies, since this technique allows to resolve the highly heterogeneous mosaic of landcover, environmental conditions, vegetation and consequently GHG fluxes that characterize large fractions of the Arctic. For all plots, general (coordinates, time of measurements, etc.), physical (pH, vegetation composition, water table, etc.) site characteristics and CH₄ and CO₂ fluxes were collected. However, for some of these parameters, data coverage turned out to be too sparse to complete analyses. For example, permafrost depth, pH, and water table level cover only 45, 15 and 34% of all available plots. To improve this situation, remotely-sensed data was included, allowing to equally cover all measurement points, albeit often with less accuracy.

Based on statistical processing using agglomerative hierarchical cluster analysis, we divided all observations into wetland categories based on their CO₂ and CH₄ flux signatures and the response to dominant environmental factors. We present the most successful classifications for different total numbers of classes, allowing to base the choice of scheme on the information that is available for a specific modelling study.