A comparison of the different methods for estimating turbulent fluxes and their errors

The near-surface gradients of wind speed, temperature and specific humidity combined with the radiative and soil heat fluxes have been widely applied in estimating the surfaceair turbulent exchange over decades. However, different methods may produce significant differences and errors in their results. Based on the analyses of the characteristics of micrometeorology in the surface layer, comparisons between the turbulent fluxes estimated by the Bowen ratio method (BRM), the two aerodynamic methods (AMs) and the variational method (VM) are presented and analyzed. A sensitive study and theoretical analyses are conducted to reveal the sources of the differences between the results from different methods and their errors. The measurements from “The Oasis System Energy and Water Cycle Field Experiment” were taken over a farmland underlying surface in the Jinta oasis during a period from June 7 to June 17, 2005. The results indicate that, （1） fluxes estimated by the BRM were strongly instable in the nighttime, thus the BRM is not applicable for estimating fluxes over oasis surface at night; (2) in comparison with the BRM, the VM could well restrain the instability for nighttime data, and show a good correlation with those from the BRM for daytime results; at the same time, the VM is found to be more accurate than the aerodynamic methods (AMs) in terms of satisfying the surface energy balance in the daytime, because of its complement of radiative and soil heat fluxes data; (3) time series of latent heat flux estimated by the VM responded more reasonably to the changes of radiation and micrometeorological conditions compared with those by the AMs; (4) the sensitivity tests show that the BRM and the AMs are sensitive to data errors, while the VM is much more stable; (5) further study indicates that the errors of the BRM could be amplified fast for -2<β<0, when the Bowen ratio is in the vicinity of -1 and that the accuracies of the AMs depend largely on their similarity functions used. In brief, the VM is found to be more rational and stable compared with the other methods. This study provides useful information for better applications of these methods.