Interpreting eddy covariance data from heterogeneous Siberian tundra: land-cover-specific methane fluxes and spatial representativeness

The non-uniform spatial integration inherent in the eddy covariance (EC) method provides an additional challenge for data interpretation when fluxes are measured in a heterogeneous environment, as the contribution of different surface types varies with flow conditions, potentially resulting in a bias as compared to the true areally averaged fluxes and surface attributes. We modelled flux footprints and characterized the spatial scale of our EC measurements at Tiksi, a tundra site in northern Siberia, including a comparison of different source area definitions. We used leaf area index (LAI) and land cover class (LCC) data, derived from very high spatial resolution satellite imagery and field surveys, and quantified the sensor location bias. We found that methane (CH 4 ) fluxes varied strongly with wind direction (from −0.09 to 0.59 μg m −2  s −1 on average), reflecting the distribution of different LCCs. Using footprint weights of grouped LCCs as explanatory variables for the measured CH 4 flux, we then developed a multiple regression model to estimate LCC-specific fluxes. This model showed that wet fen and graminoid tundra patches in locations with a high topography-based wetness index acted as strong CH4 sources (0.95 μg m −2  s −1 ), while mineral soils were significant sinks (−0.13 μg m −2  s −1 ). Finally, to assess the representativeness of CH 4 flux measurements, we upscaled the LCC-specific fluxes to different spatial scales. This assessment showed that, despite the surface heterogeneity and rather poor representativeness of EC data with respect to the areally averaged LAI and coverage of some LCCs, the mean CH 4 flux measured during summer 2014 was close to the corresponding balance upscaled to an area of 6.3 km 2 , with a location bias of 14 %. We recommend that EC site descriptions in a heterogeneous environment should be complemented with footprint-weighted high-resolution data on vegetation and other relevant site characteristics.