Model-based Source Partitioning of Eddy Covariance Flux Measurements

Abstract

Terrestrial ecosystems constantly exchange momentum, energy, and mass (e.g., water vapor, CO₂) with the atmosphere above. This exchange is commonly measured with a micrometeorological technique, the eddy covariance (EC) method. Various components of the measured net fluxes, such as transpiration, evaporation, gross primary production, and soil respiration, cannot be depicted separately by the EC approach. Thus, so-called source partitioning approaches have to be applied to CO₂ and water vapor EC data to gain a better understanding of the prevailing processes and their interrelations in terrestrial ecosystems. A large variety of partitioning procedures with diverse model approaches have been developed, including various driving variables, necessity of different input data and parameterizations. The most robust and commonly used source partitioning tools for CO₂ flux components, often primarily developed to fill gaps in EC measurements, are based on the notion that during night respiration fluxes prevail. They use non-linear regressed relationships of these nighttime observations and physical drivers (e.g., temperature in the approach after Reichstein et al. 2005). Here, the challenge lies within extrapolating the nighttime relationship to daytime conditions, and analogous methods for water fluxes are lacking. In this thesis, next to the approach after Reichstein et al. (2005) various data-driven source partitioning approaches for H₂O and CO₂ fluxes were applied, compared, modified, and evaluated for multiple ecosystems to get a better understanding of the methods’ functionality, dependencies, uncertainties, advantages, and shortcomings.<br /> We first describe the coupling and extension of the complex terrestrial ecosystem model AgroC. Further, we conducted a comprehensive model-data fusion study to clarify the CO₂ exchange in agroecosystems and estimate their annual carbon balance. For three test sites in Western Germany, AgroC was calibrated based on soil water content, soil temperature, biometric, and soil respiration measurements for each site, and validated sufficiently in terms of hourly net ecosystem exchange (NEE) measured with the EC technique. Moreover, AgroC reproduced the flux dynamics very effectively after sudden changes in the grassland canopy due to mowing. In a second step, AgroC was optimized with the EC measurements to examine the effect of various objective functions, constraints, and data-transformations on the estimated carbon balance and to compare the results to the established gap-filling approach after Reichstein et al. (2005). It was found that modeled NEE showed a distinct sensitivity to the choice of objective function and the inclusion of soil respiration data in the optimization process. Even though the model performance of the selected optimization strategies did not diverge substantially, the resulting cumulative NEE over simulation time period differed extensively. Therefore, it is concluded that data-transformations, definitions of objective functions, and data sources have to be considered cautiously when a terrestrial ecosystem model is used to determine NEE by means of EC measurements.<br /> Second, we applied the source partitioning approaches after Scanlon and Kustas (2010; SK10) and after Thomas et al. (2008; TH08) to high frequency EC measurements estimating transpiration, evaporation, net primary production, and soil respiration, of various ecosystems (croplands, grasslands, and forests). Both partitioning methods are based on higher-order statistics of the H₂O and CO₂ fluctuations, but proceed differently. SK10 had the tendency to overestimate and TH08 to underestimate soil flux components, where the partitioning of CO₂ fluxes was more irregular than of H₂O fluxes. Results derived with SK10 showed relatively large dependencies on estimated water use efficiency (WUE) on leaf-level, which is needed as an input. Measurements of outgoing longwave radiation used for the estimation of foliage temperature and WUE could slightly increase the quality of the partitioning results. A modification of the TH08 approach, by applying a cluster analysis for the conditional sampling of respiration/evaporation events, performed sufficiently, but did not result in significant advantages compared to the other method versions. The performance of each partitioning approach was dependent on meteorological conditions, plant development, canopy height, canopy density, and measurement height. Foremost, the performance of SK10 correlated negatively with the ratio between measurement and canopy height. The performance of TH08 was more dependent on canopy height and leaf area index. It was found, that all site characteristics which increase dissimilarities between scalars enhance partitioning performance for SK10 and TH08.<br /> Also, we conducted large eddy simulations (LES), simulating the turbulent transport of H₂O and CO₂. SK10 was applied to the synthetic high frequency data generated by LES, and the effects of canopy type, measurement height, given scalar sink-source-distributions, and estimated WUE input were tested regarding the partitioning performance. The LES-based analysis revealed that for a satisfying performance of SK10, a certain degree of decorrelation of the H₂O and CO₂ fluctuations was needed and a correct WUE estimation was favorable. Furthermore, another possible error source, which was so far not yet discussed in the literature, could be detected for the partitioning approach. In the special case of the LES experiments, validity of an essential assumption about the prevailing transport efficiencies of the scalars in the method’s derivation was found to be a crucial point for a correct application of SK10.<br /> The application of different source partitioning methods including their various involved assumptions, required input data and work effort showed that still uncertainties and unknowns prevail for the source partitioning of water vapor and CO₂ fluxes. An assessment and evaluation of the partitioning results can only be conducted with additional measurements of flux components on differing spatial and temporal scales independent of the EC measurements. Further, the application of multiple partitioning methods (usage of an ensemble) to the same data can give a better idea about uncertainties in the results.