Comparing eddy covariance fluxes of carbon dioxide, latent and sensible heat for Dublin using TK3, EddyPro and EddyUH

There is a scarcity of published eddy covariance (EC) software comparisons despite the increasing number of software packages available to the international EC community. This paper presents results for fluxes of carbon dioxide (FC), latent (QE) and sensible heat (QH) from an EC software comparison using a traditional, ‘tried and tested' software, Turbulence Knight (TK3) developed by the Micrometeorology Department of the University of Bayreuth with 2 novel software 1) EddyPro developed by Licor Biosciences and 2) EddyUH developed by the Department of Physics, University of Helsinki. A three month EC dataset from Dublin City is used and a consistent post-field data processing scheme is employed in all 3 software.

The instruments deployed at the Dublin City EC site include a Campbell Scientific sonic anemometer (CSAT3) and a Licor open path (OP) gas analyser (Li7500). As a consequence of deploying an OP gas analyser periods when precipitation are recorded on site are omitted from the final results, in addition to periods corresponding to diagnostic variables that indicate instrumental problems. Large differences were found among the software with regard to the number of values deemed as bad quality or NANs. Despite implementing the Vickers and Mahrt (1997) de-spiking algorithms in EddyPro and EddyUH it was found that different interpretations of this algorithm lead to differences in the enumeration of spikes and hence the final available output. TK3 implements a novel de-spiking algorithm employing the median absolute deviation (MAD) and also resulted in a different enumeration and determination of spikes.

In order to minimise the effect of the different de-spiking algorithms on the results only 30 minute intervals when all three software reported flux estimates were considered for the comparison. Following the elimination of periods reporting precipitation events, instrumental problems and bad values (NANs), the results illustrate good agreement across all three software for fluxes of QE, QH and FC. Greater disagreement is reported in terms of quality flag agreement, and is likely a result of differing implementations of the flagging criteria in each of the software.Overall the research shows that all 3 software are suitable for use in urban locations, like Dublin, where OP gas analysers are deployed, however greater flexibility in terms of coordinate rotation methods (i.e. sector-wise planar fit method) and spectral corrections (i.e. in-situ derived methods, which are optimal when closed-path analysers are deployed) is provided in EddyPro and EddyUH.