 |
| Titel |
Failure of correct evapotranspiration measurements by eddy covariance under certain conditions and energy balance closure in open-oak savanna ecosystems |
| VerfasserIn |
Oscar Perez-Priego, Mirco Migliavacca, Tarek El-Madany, Arnaud Carrara, Gerardo Moreno, Olaf Kolle, Markus Reichstein |
| Konferenz |
EGU General Assembly 2016
|
| Medientyp |
Artikel
|
| Sprache |
en
|
| Digitales Dokument |
PDF |
| Erschienen |
In: GRA - Volume 18 (2016) |
| Datensatznummer |
250134129
|
| Publikation (Nr.) |
 EGU/EGU2016-14823.pdf |
|
|
|
|
|
| Zusammenfassung |
| Separation of evapotranspiration (ET) into its components represents one of the main
ecohydrological challenges in heterogeneous ecosystems (i.e. tree-grass savanna),
where two main evaporative layers consisting of tree canopy (ETabove) and its
underlying surface (ETsubcanopy) dominate ET. The challenge arises from the fact that
classical eddy covariance 1) directly only measures total ET and 2) biases in the
respective energy balance are often observed. Here, we address these challenges in a
Mediterranean savannah tree-grass ecosystem, by synchronous, combined measurements via
classical eddy covariance, sub-canopy eddy covariance, sap-flow, and replicated
lysimeters.
To this end, half-hourly latent heat fluxes of the grass layer estimated using six novel
lower boundary-tension and -temperature controlled lysimeters (LEsubcanopy−lysimeter)were
compared to those measured by a sub-canopy eddy covariance tower placed at
1.8 m (LEsubcanopy−eddy) over a year. To explain the residuals (epsilon) between
LEsubcanopy−lysimeter and LEsubcanopy−eddy , we trained a random forest model (RF) using
soil moisture (SM), ground-heat fluxes (G), net radiation (Rn), air relative humidity (RH) and
friction velocity (u*) as main predictor variables. The degree of energy closure was evaluated
by comparing residual LE (LEresidual, estimated as Rn-H-G; H denotes sensible heat
flux) against total LE measured by a tall tower installed above the canopy at 15 m
(LEeddy). In parallel, we contrasted this using independent, upscaled LE (LEupscaled=
LEsubcanopy−lysimeter + LEabove−sapflow; being LEabove−sapflow the tree component
derived from sap-flow measurements) to test whether failures in LEeddy explain the lack
of energy balance closure. In such a case, we test the use of RF as a generalized
approach to estimate epsilon and correct for LEeddy (LEeddy−corrected = LEeddy +
epsilon).
As main results, the comparison of independent LEsubcanopy−eddy and
LEsubcanopy−lysimeter evidenced that eddy covariance tends to underestimate LE
under certain conditions. To assess the relative importance of the most important
predictor variables, a permutation-based test was carried out. In order of importance,
SM and G showed to be the most determinant in predicting epsilon in our site.
More in details, results from a partial dependence analysis showed that highest
disagreements hold at high SM, G, Rn and RH and low u*. Errors in LEeddy explained
the lack of closure in the surface energy balance (slope=0.79, R2=0.64). In fact,
a better agreement was obtained with LEupscaled(slope=0.92, R2=0.70), which
support such hypothesis. Consistently, an even better result was obtained by using
LEeddy−corrected (slope=1.01, R2=0.70) and points over the generalization of the RF model
to correct for LEeddy. Promising results were also obtained by predicting epsilon
in similar sites with similar instrumental set-up. The implication of such results,
together with the importance of other source of uncertainties (i.e. spatial and temporal
cross-scale issues among methods, footprint analyses, and surface heterogeneity) in
determining the degree of closure in the surface energy balance will be further discussed. |
|
|
|
|
|
|