Environmental Engineering Reference
In-Depth Information
of the large variability of ecosystems traits of sites that are included in this PFT;
in fact the DBF sites in the networks are highly heterogeneous, with mature for-
ests (IT-Col), coppices with different ages (IT-Ro1 and IT-Ro2) and a poplar plan-
tation (IT-PT1). PFT annual NEE sensitivity to CUP variations range between
−
2.97 0.64 for CRO and
−
3.80 0.56 C m
−
2
yr
−
1
day
−
1
for GRA.
Within some PFTs, sites with particular behaviors emerged (e.g. IT-Cpz for
EBF and IT-Pia, included in this analysis in the grassland PFT) probably because
of the effect of some site properties (in IT-Cpz for example there is the effect of
the water table depth) or ecological parameters (e.g. LAI, fertility, …) on the rela-
tion between annual NEE and CUP. These effects generally resulted in a similar
NEE sensitivity to CUP variations but with different absolute values.
Moreover CRO and GRA, PFTs where management can play an important role,
showed quite good relationship between NEE and CUP and sensitivity values sim-
ilar to those obtained in the other PFTs (Fig.
2.5
a, e).
2.4.3 Analysis of the Relationship Between Annual NEE and
Climate Variables
The analysis of the relationship between annual cumulated NEE and climate vari-
ables has been conducted by aggregating site years according to the PTF. The
analysis has been conducted in three different steps:
1. As a first attempt, a correlation analysis between annual NEE and average
annual climate variables has been conducted. As climate variables temperature,
precipitation, radiation and vapor pressure deficit (VPD) have been used.
2. In a second step, the stepwise regression has been used for the selection of the best
predictive variables of annual NEE. The predictive variables tested were meteoro-
logical variables (T, Prec, VPD, Rg), LAI- and phenological indicators (i.e. CUP).
A stepwise approach based on the Akaike's Information Criterion (AIC) for model
selection has been implemented. The AIC is a measure of the trade-off between
the goodness-of-fit (model explanatory power) and model complexity (number of
parameters). Therefore, the stepwise based on AIC is a multiple regression method
for variable selection which accounts on one hand for the model explanatory
power, and on the other for the increasing complexity of the model when addi-
tional variables are tested in the multiple regression model (Venables and Ripley
2002
; Yamashita et al.
2007
). The stepwise AIC was preferred to other stepwise
methods for variable selection since it can be applied to non-normally distributed
data (Yamashita et al.
2007
). The data have been weighted according to the uncer-
tainty estimates (NEE
max-min
), defined as described in Sect.
2.4.1
The outcome of
the stepwise AIC is a multiple linear model including the best set of predictive
variables explaining the variability of annual NEE. Therefore, by applying step-
wise AIC as explained above, the drivers explaining the variability of NEE for
each PFT have been identified while the coefficients estimated for each selected
variable represent the sensitivity of NEE to variations of each predictor.
