Assessing and correcting spatial representativeness of tower eddy-covariance flux measurements
Abstract:Estimating the landscape-scale exchange of ecologically relevant trace gas and energy fluxes from tower eddy-covariance (EC) measurements is often complicated by surface heterogeneity. For example, a tower EC measurement may represent less than 1% of a grid cell resolved by mechanistic models (order 100–1000 km2). In particular for data assimilation or comparison with large-scale observations, it is hence critical to assess and correct the spatial representativeness of tower EC measurements.
We present a procedure that determines from a single EC tower the spatio-temporally explicit flux field of its surrounding. The underlying principle is to extract the relationship between biophysical drivers and ecological responses from measurements under varying environmental conditions. For this purpose, high-frequency EC flux processing and source area calculations (≈60 h−1) are combined with remote sensing retrievals of land surface properties and subsequent machine learning. Methodological details are provided in our companion presentation “Towards the spatial rectification of tower-based eddy-covariance flux observations”.
We apply the procedure to one year of data from each of four AmeriFlux sites under different climate and ecological environments: Lost Creek shrub fen wetland, Niwot Ridge subalpine conifer, Park Falls mixed forest, and Santa Rita mesquite savanna. We find that heat fluxes from the Park Falls 122-m-high EC measurement and from a surrounding 100 km2 target area differ up to 100 W m−2, or 65%. Moreover, 85% and 24% of the EC flux observations are adequate surrogates of the mean surface-atmosphere exchange and its spatial variability across a 900 km2 target area, respectively, at 5% significance and 80% representativeness levels. Alternatively, the resulting flux grids can be summarized as probability density functions, and used to inform mechanistic models directly with the mean flux value and its spatial variability across a model grid cell. Lastly, for each site we evaluate the applicability of the procedure based on a full bottom-up uncertainty budget.