Fig. 1.
Location of sites in France.
Table 1.
Experimental sites characteristics.
03°35’45”E,
270 m elevationPFT: Evergreen Broadleaf Forest (EBF)
Species:
Holm oak (Quercus ilex)
Understorey : Buxus sempervirens, Phyllirea latifolia, Pistacia terebinthus and Juniperus oxycedrus
LAI: 2.7 ± 0.3 m2 m-2 (overstorey)
Management:
Coppice
Last cut : 1942Climate :
Mediterranean
Soil:
Shallow bedrock :
75% in the top 0–50 cm 90% below
Stone-free fine fraction of the top-soil:
38.8% clay
35.2% silt
26% sand.
Water holding capacity: 140 mm.Period: 1984–2016
Temperature: 13.3 °C
Precipitation: 923 mmAllard et al. (2008);
Cabon et al. (2018)BarbeauFR-Fon48°28’37’’N,
02°46’56’’E,
89 m elevationPFT: Deciduous Broadleaf Forest (DBF)
Species :
Dominant : sessile oak (Quercus Petraea, (Matt.) Liebl. 160 yrs old)
Sublayer : Hornbeam (Carpinus betulus L., 60-70 yrs old)
LAI : 5.3 m2 m-2 (2005–2016 average)
Management : coppiceClimate:
Mild oceanic temperate
Soil:
Oligo-mull humus,
Gleyic luvisol:
30% clay
37% silt
33% sand
Millstone bedrock at 0.9 m depth
Water holding capacity: ∼ 170 mm.Period: 1980–2010
Temperature: 11.2 °C
Precipitation: 677 mmDelpierre et al. (2009); Delpierre et al. (2016)Le BrayFR-LBr44°43’02”N,
00°46’10”W,
62 m elevationPFT: Evergreen Needleleaf Forest (ENF)
Species:
Mature maritime pine (Pinus Pinaster Aït.)
Understorey : Molench coerulea (L.) Moench
LAI : 2.6 to 3.1 m2 m-2 (overstorey)
Management:
Seeding: 1970
Thinning: 1990, 1995, 2002, 2005Climate:
Oceanic temperate
Soil:
Humic sandstone :
2% clay
98% sand
Impermeable layer of Ferruginous cement in the B horizon at 0.9 m depth (alios).
Water holding capacity: ∼120 mm.Period: 1987–2008
Temperature: 12.5 °C
Precipitation: 932 mmLoustau and Granier (1993);
Granier and Loustau (1994);
Jarosz et al. (2008);
Ogée et al. (2003)AuradéFR-Aur43°32’59’’N,
01°06’22’’E,
245 m elevationPFT: Cropland (CRO)
Species:
Rotation of rapeseed, winter wheat and sunflower (see Table S1)
PAI : 0 to 5 m2 m-2 (2005–2014)
Management : No irrigation (see Table S1)Climate:
Temperate
Soil:
32.3% clay
47.1% silt
20.6% sand,Period: 2004–2014
Temperature: 13 °C
Precipitation: 680 mmBéziat et al. (2009); Etchanchu et al. (2017)LaqueuilleFR-Laq45°38’34’’N,
02°44’02’’E,
1040 m elevationPFT: Extensive grassland (GRA)
Species: Grass
LAI : 1.5 m2 m-2 (average)
Management: Extensive
no fertilization and grazed continuously by heifers from May to October with a lowLivestock Unit (LSU) of 0.6 SLU ha-1 yr-1Climate:
Temperate/continental
Soil:
Andosol developed on basaltic rocks.
Loamy texture
content (18% with a C:N ratio of 10.5)Period: 2004–2014
Temperature: 7.9 °C
Precipitation: 1000 mmKlumpp et al. (2011)
Table 2.
Setup at each site. Tair stands for air temperature at measurement height (in °C), RH for relative humidity (in %), Sd for incoming shortwave radiation (in W m−2), u for wind speed (in m s−1), P for precipitation (in mm), θ for soil water content (in mm).
Measurement heightData processedVariablesInstrumentsStatistical analysis period (1 or 2)Puechabon 12.2m2001–2014Meteorological variables :Tair/RH :
Sd :
u :
P :
θ :MP100 (Rotronic Inst. Corp., Huntington, NY, USA)
SKS1110 (Skye Inst. Ltd, Powys, UK)
A100R (Campbell Scientific Inc, Logan, UT, USA)
ARG100 (Environmental Measurements Ltd., Sunderland, UK)
TDR Trase (Soil Moisture Equipment Corp., Santa Barbara, CA, USA)2001–20141
2006–20112Micrometeorological variables :3D u :
CO2/H2O :GILL-R3 (Gill Instruments Ltd, Lymington, Hampshire, UK)
LI-6262 (Li-Cor Inc., Lincoln, NE, USA)Barbeau 37m2005–2018Meteorological variables :Tair/RH :
Sd :
u :
P :HMP155A (Vaisala, Helsinki, Finland)
CMP22 (Kipp & Zonen, Delft, NL)
WindSonic (Gill Instruments Ltd)
3029/2 (Précis Mécanique SAS, Bezons, France)2005–20181
2006–20112Micrometeorological variables :3D Ws : 2005-2018
2012-2018
CO2/H2O : 2005-2018
2012-2018GILL-R3-50 (Gill Instruments Ltd)
GILL-HS-50 (Gill Instruments Ltd)
LI-7500 (Li-Cor Inc., Lincoln, NE, USA)
LI-7200 (Li-Cor Inc., Lincoln, NE, USA)Le Bray 41m2000–2008Meteorological variables :Tair/RH :
Sd :
u :
P :
θ :HMP45 (Vaisala, Helsinki, Finland)
CE180 (Cimel Electronique, Paris, France)
05103 (Young, Traverse city, MI, USA)
ARG100 (Campbell Scientific Inc, Logan, UT, USA)
TDR Trase (Soil Moisture Equipment Corp., Santa Barbara, CA, USA)2000–20081Micrometeorological variables :3D u :
CO2/H2O :GILL-R2 (Gill Instruments)
LI-7500 (Li-Cor Inc., Lincoln, NE, US)Auradé 2m2005–2014Meteorological variables :Tair/RH :
Sd :
u :
P :
θ :HMP35A (Vaisala, Helsinki, Finland)
CNR1/CNR4 (Kipp & Zonen, Delft, NL)
05103 (Young, Traverse city, MI, USA)
ARG100 (Environmental Measurements Ltd., Sunderland, UK)
CS616 (Campbell Scientific Inc, Logan, UT, USA)2005–20141
2006–20112Micrometeorological variables :3D u :
CO2/H2O :CSAT3 (Campbell Scientific Inc, Logan, UT, USA)
LI-7500 (Li-Cor Inc., Lincoln, NE, USA)Laqueuille 2m2004–2013Meteorological variables :Tair/RH :
Sd :
u :
P :
θ :HMP155A (Vaisala, Helsinki, Finland)
SKS1110 (Skye Inst. Ltd, Powys, UK)
05103 (Young, Traverse city, MI, USA)
52202 (Young, Traverse city, MI, USA)
CS616 (Campbell Scientific Inc, Logan, UT, USA)2004–20131Micrometeorological variables :3D u :
CO2/H2O :GILL-R3 (Gill Instruments Ltd)
LI-7500 (Li-Cor Inc., Lincoln, NE, USA)
Fig. 2.
RF analysis results: importance of environmental variables explaining the variability of carbon fluxes (A. NEE, B. GPP and C. Reco) at a half-hourly time-step at five ecosystem stations. Standard deviation on the mean of %imp of each environmental variable, computed from a bootstrap analysis (n = 100), were voluntarily omitted in the graph as they presented values lower than 0.1% in each case (See Supplementary material S3.1).
Table 3.
Input parameters optimization and results of the RF approach in terms of percentage of variance explained from the half-hourly (five sites), daily, monthly and seasonal (two sites selected) time scales. « ndata » corresponds to the number of data in the time-step considered; « mtry » stands for the number of explanatory variables to randomly sample as candidates at each split and « ntrees » for the number of trees in the RF analysis.
See Supplementary material (§ S3.1) for further information.
Fig. 3.
RF analysis results: importance of environmental variables explaining the carbon flux variabilities for Puechabon (A) and Auradé (B) at different time scales. Standard deviation on the mean of %imp of each environmental variable, computed from a bootstrap analysis (n = 1000), were voluntarily omitted in the graph as they presented values lower than 1.5% for each case (§ S3.1).
Fig. 4.
2006–2011 monthly time series of environmental variables (Sd, Tair, Cumulative P, u and D) for the three sites selected in the wavelet analyses.
Fig. 5.
2006–2011 monthly time series of carbon fluxes (NEE, GPP, and Reco), for the three sites selected in the wavelet analyses.
Fig. 6.
Wavelet Power Spectra (WPS) of environmental variables (scalograms) and corresponding profile of the average wavelet power (curves) at three sites (Puechabon, Barbeau and Auradé).
Fig. 7.
Wavelet Power Spectra (WPS) of ecosystem carbon fluxes (scalograms) and corresponding profile of the average wavelet power (curves) at three sites (Puechabon, Barbeau and Auradé).
Fig. 8.
Cross-wavelet coherence (CWC) analysis of GPP (B) and Reco (C) with D at three sites (Puechabon, Barbeau and Auradé) from 2006 to 2011. The monthly time series of variables (A) are presented together with the corresponding scalograms of CWC. The red colour indicates high coherence between the two-time series at a particular time and a particular period. A blue one indicates no coherence between the two time series. Arrows on the cross-wavelet coherence plots stand for the synchronization analyses and are plotted only within white contour lines indicating significance (with respect to the null hypothesis of white noise processes) at the 5% level. Arrows are pointing right for the in-phase two series relationship (e.g., positively correlated with no lags), left for the anti-phase relationship (e.g., negatively correlated with no lags), and straight up (down) when carbon flux (biometeorological variable) is leading the biometeorological variable (carbon flux) by π/2 radians. Finally, the grey zone represents the cone of influence in which the spectral information is likely to be less accurate (Torrence & Compo 1998).
Fig. 9.
Cross-wavelet coherence (CWC) analysis of GPP (B) and Reco (C) with θEW at three sites (Puechabon, Barbeau and Auradé) from 2006 to 2011. The monthly time series of variables (A) are presented together with the corresponding scalograms of CWC.