Use of the daily differencing approach to evaluate uncertainties affecting eddy covariance measurements

The eddy covariance technique is recognised to be the most adapted micrometeorological
method to study the exchange processes between terrestrial ecosystems and
the atmosphere. Like all other methods, it is submitted to systematic and random measurement
errors. A thorough analysis of these errors is necessary in order to set the
limits of validity of the method and to quantify the uncertainty that affects net carbon
exchange computed with this method.
In this presentation, we’ll concentrate on the random errors using the dailydifferencing
approach (DDA) developed by Hollinger and Richardson (2005). The
interest of this approach is that it requests only single tower measurements and is
thus applicable to all flux tower sites. In this approach, uncertainties are estimated by
comparing flux values taken at two successive days at the same hour and under similar
meteorological conditions. The analysis may be applied to sensible heat, latent
heat and CO2 flux densities. It was applied here to the eddy-covariance data from the
Vielsalm mixed forest site (10 years of data) and from the Lonzée agricultural site (4
years of data). Both sites are situated in Belgium and are part of the Carboeurope IP
network. The study is developed in the frame of the European IMECC project.
The absolute and relative random error was quantified for both sites. Their daily evolution
and their dependencies on different climate conditions (magnitude of the flux,
PPFD, Rnet, wind velocity, wind direction, clarity index) were analysed. For both
sites, the absolute random error increases linearly with the absolute value of flux. This
is the principal factor controlling the random error. More particularly the response of
the random error to wind velocity was analysed. For the CO2 flux, the absolute random
error decreases with increasing wind speed. This effect is more important for the
agricultural site than for the forest site. The behaviour of the relative random error
with wind speed is more contrasted: it generally decreases with increasing wind speed
at low velocities but, for some directions may increase with wind speed at high velocities.
In addition, the random error was found very dependent on wind direction at the
forest site probably as a result of site heterogeneity.