PhD Thesis
Adoption of footprint methods for the quality control of eddy-covariance measurements
Mathias Göckede (10/2001-05/2005)
Support: Thomas Foken
Footprint models determine the spatial context of a measurement by defining a transfer function between sources or sinks of the signal and the sensor position. The resulting source area provides an important quality control tool to improve the interpretation of micrometeorological data sets, e.g. by assessing the influence of distorting terrain elements on the measurements. However, to date no approaches have been presented in the literature that provide a standardised footprint-based methodology that allows observers to include terrain characteristics into quality assessment and quality control strategies. Consequently, it has not yet been possible to conduct studies comparing the sites organised in flux monitoring networks such as FLUXNET (e.g. Baldocchi et al., 2001) while taking into account the influence of the local terrain structure on the data quality. One problem in this context is the small number of studies that concentrate on the validation of footprint models under the non-ideal conditions in which they are frequently being used (Foken and Leclerc, 2004). Therefore, for many applications, e.g. in aerodynamically inhomogeneous terrain, the accuracy of the source areas computed by the footprint models cannot be evaluated. To further increase the acceptance of footprint-based studies, a stronger focus on footprint validation studies for a wide variety of experimental designs is needed.
This dissertation focuses on the development of a footprint-based evaluation tool for complex measurement sites that allows the combination of quality assessment results for micrometeorological measurements with characteristics of the surrounding terrain. The standardised method is easy-to-use in order to encourage its application on a large number of sites. To improve the interpretation of the obtained results, a second objective of this thesis was to develop and test approaches to validation experiments for footprint models. In this context, several studies on natural tracer experiments for footprint validation purposes as a low-cost and practical alternative to footprint validation experiments using artificial trace gases were performed.
Göckede et al. (2004) presented an approach for the evaluation of micrometeorological measurement sites in complex terrain, which combined a method for quality assessment of eddy-covariance measurements (Foken and Wichura, 1996) with an analytic footprint model (Schmid, 1994; 1997). Their software package provided micrometeorologists, for the first time ever, a practical tool for determining the average flux contributions from the land use type intended to observe at a specific site, or to identify footprint areas for which a high data quality could be assumed, to name some examples. Rebmann et al. (2005) proved the efficiency of this evaluation approach for extensive studies on a large number of sites organised in a network by comparing 18 different sites of the CARBOEUROFLUX project. Although the average data quality for the sites tested was high, they were able to demonstrate negative effects of surface heterogeneity on the average flux data quality, and also problems caused by the instrumentation itself, such as misalignment of the sensor or flow distortion by the tower. These results may serve as a tool for an improved determination of yearly sums of the net ecosystem exchange, because fluxes originating from sectors of minor quality could be excluded from the analysis. Because of these important contributions to quality control, Foken et al. (2004) integrated the site evaluation approach into a comprehensive survey on micrometeorological post-field data quality control techniques. The experiences obtained during the extensive study by Rebmann et al. (2005) allowed us identification of the major weak points of the original site evaluation approach by Göckede et al. (2004), which we were able to improve in subsequent studies. Using remote sensing methods Reithmaier et al. (2005) studied the influence of the characteristics of the land use maps and different roughness length assignment schemes on the performance of the site evaluation approach. Finally, Göckede et al. (2005a) developed an updated version of the site evaluation approach, which improved the basic method by replacing the analytic footprint model with a Lagrangian stochastic footprint model (Rannik et al., 2003) that is more suitable for studies above high vegetation, and by applying a more sophisticated microscale flux aggregation method (Hasager and Jensen, 1999) for the determination of areally-averaged roughness lengths. This software package forms an optimum compromise between the accuracy of the modelling results and an easy applicability to various sites. Although the implemented models are far more sophisticated than in the original version, the approach by Göckede et al. (2005a) still permits a practical application that allows for comparative studies of a large number of sites. A further improvement of the remaining conceptual weak points, such as the assumption of horizontally homogeneous flow conditions by the employed forward LS footprint model, would require extensive input data sets which could only be provided for detailed analyses of single selected study sites.
With respect to the development of validation methods for footprint models using natural tracer measurements from field scale experiments, Göckede et al. (2005b) presented two different experimental approaches. The first of these, a comparison of measured flux differences and modelled land use differences for pairs of measurement positions, revealed general correlations between measurement data and model results. However, a definite equation for a correlation analysis between flux measurements and source area composition could not be identified and, as a consequence, a quantitative evaluation of the results was not possible. Secondly, Göckede et al. (2005b) tested a correlation analysis between measured and modelled parameters using reference measurements and footprint results. Due to a clearly linear functional relationship between measured and modelled quantities, this approach resulted in an objective quantitative evaluation of the accuracy of the footprint model. The study by Reth et al. (2005), which among other objectives attempted to use soil chamber measurements and eddy-covariance data to evaluate footprint models, could not be employed for footprint validation purposes because of a large systemic scatter between these measurement systems. Overall, both the paper by Göckede et al. (2005b) and by Reth et al. (2005) provided successful methods to testing the suitability of natural tracer experiments in the validation of footprint models. Although experimental deficits prevented the working out of significant differences between the results of the employed footprint models, their studies developed an improved design for natural tracer experiments that are especially designed for footprint validation purposes. These results could form the basis for future experiments that may improve the application of footprint models in non-ideal terrain conditions.