Post on 01-Jan-2016
description
NEW TRANSFER FUNCTIONS FOR CORRECTING EDDY NEW TRANSFER FUNCTIONS FOR CORRECTING EDDY COVARIANCE FLUXES OF WATER VAPOURCOVARIANCE FLUXES OF WATER VAPOUR
Unit of Biosystem Physics – Gembloux Agricultural University – Belgium
A. De Ligne, B. Heinesch, M. Aubinet
Unité de Physique des Biosystèmes, Faculté Universitaire des Sciences Agronomiques de Gembloux, Belgium
Poster presented at the IMECC Annual meeting in Geneva (Switzerland), the 28th January 2009.
To find transfer functions that are more appropriate to water vapour fluxes.
To study dependencies of the transfer function parameters to air humidity.
To calibrate and validate a modelled transfer function on measurements performed at a forested site (Vielsalm, Belgium).
SITE DESCRIPTION:
METHODS
Experimental TF
OBJECTIVES
Vielsalm, Belgium (50°18 N, 6°00 E) : Mixed forest site, Temperate maritime climate, 13 years of data.
Turbulent water vapor fluxes measured with closed-path eddy covariance systems are affected by high frequency fluctuation attenuation.
The classical transfer functions used to correct this error (Gaussian and Lorentzian) are not appropriate because :
-They do not take into account the attenuation dependency on air humidity
- Their shape is not adapted to the observed transfer functions.
INTRODUCTION
USUAL TF EQUATIONSa) Gaussian equation
(Aubinet et al, 2001)
f : frequency
fc : cut-off frequency (frequency for which TF = ½)
b) Lorentzian equation,
(Eugster and Senn, 1995)
NEW TF EQUATIONSc) Reduced slope TF
d) Double TF
Similar as the Gaussian but with a smaller exponent x
Two adjustable parameters : fc and x.
x
cCO f
f
f
ffTF 2ln2lnexp
2
2
x
cf
ffTF 2lnexp
2
1
1
cff
fTF
Product of the CO2 TF (passive tracer) and of an additional TF describing adsorption / desorption.
The additional TF has another cut-off frequency and exponent.
Two adjustable parameters : fc and x.fig 1: Representation of the four transfer function equations adjusted on a same sample. The dotted lines
in b) c) and d) are the Gaussian equation a) taken as reference.
Transfer function (TF) equations
TRANSFER FUNCTION RESPONSE TO VAPOUR PRESSURE DEFICIT (hPa)
PERSPECTIVES
Ds (hPa)
a) Gaussian
b) Lorentzian
c) Reduced slope
d) Double TF
Number of samples
[0 – 5] 2.64 2.02 1.89 2.27 205
[5 – 10] 2.21 1.73 1.40 1.44 422
[10-15] 2.10 1.64 1.20 1.19 243
[15-20] 2.14 1.63 0.98 0.92 94
For Different VPD classes :
Different cut-off frequencies BUT ALSO : Different TF slopes
NB : Selection of samples according to Mammarella et al :- 6 consecutive half hours - H and LE > 25 Wm-2 and CO2 flux > 2 μmol m-2 s-1
Table 1: Sum of squared difference means (SSD) between experimental and modelled TF ranged by Ds categories.
Legend : highest SSD – lowest SSD – no significant difference
Modelled TF
2
2lnexpcf
ffTF
fc
fc
x
fc
fig 3: Families of modelled TF for different VPD classes
All models take account of the cut-off frequency increase with VPD
Models c and d take in addition account of an exponent decrease with VPD
• To improve adjustment at low VPD.
• To study the effect on correction factors.
• To extend analysis on other sites.
Fig 6: Average of experimental TF and modelled TF by Ds categories
Validation statistics
New TF equations fit better the experimental transfer functions at all VPD classes.
The two models are not significantly different at large VPD.
At low VPD (0 – 5 hPa) the adjustment is perfectible.
Re-evaluation of correction factor :
Contact : deligne.a@fsagx.ac.be
Acknowledgements: This research is funded by European Commission (IMECC project) and Belgian Federal Government (IMPECVOC project).
RESULTS OF CALIBRATION / VALIDATION
f : frequency
fc: :cut-off frequency (frequency for which TF = ½)
fig 2: Transfer functions for water vapour flux at different VPD classes
fig 4: Families of experimental (points) and modelled (lines) TF adjusted on different VPD classes
Ds (hPa) a) Gaussian b) Lorentzian c) Reduced slope d) Double TF
[0 – 5]
[5 – 10]
[10-15]
[15-20]
Table 2: Correction factor means ranged by Ds categories
It is likely that the correction factors computed with new TF will be larger.
The correction factor decreases with VPD.
The computation is in progress.