Model fit to the different data streams

You can find below the model fit to all data streams used within CARBONES CCDAS. The objective is to show the main model performances with respect to the different data steams that were assimilated. Note that these fit correspond to a “stepwise” approach where the different data streams are assimilated in sequential order, starting with MODIS-NDVI, then FluxNet data, and finally atmospheric CO2 concentrations (see CCDAS description). The ocean surface pCO2 data are used in a separate step to construct a prior ocean flux product. The sections below illustrate the model-data fit to:

1. MODIS – NDVI data
2. Net ecosystem CO2 and water fluxes (FluxNet)
3. Surface ocean partial pressure of CO2 (pCO2)
4. Atmospheric CO2 concentration at surface sites

1. Fit to MODIS – NDVI data

This section illustrates the results of step 1 of the “stepwise optimization” approach using MODIS-NDVI data. Note that only the normalized values of MODIS-NDVI are used in order to extract only the seasonality of the satellite signal. The normalized simulated fAPAR from the model is optimized against the normalized MODIS-NDVI.

Go to Top


1.1 Site locations

The figure below shows the locations of the sites used in the MODIS-NDVI optimizations of the phenology-related parameters for deciduous forests and crop Plant Functional Types. Only few specific points, with a fraction of the considered vegetation type in the MODIS pixel above 70%, are considered in the optimization. The remaining points are kept for evaluation.

 

Go to Top


1.2 Results per Plant Functional Types

Results are presented for all Plant Functional Types (PFTs) with first the PFTs where the optimization has performed relatively well, followed by results from PFTs were the optimization did not result in significant improvement to the model output. All graphics present:

• The MODIS-NDVI measurements in black.
• The Multiple-Site inversion (MS) results in blue that corresponds to the optimisation of all sites at once for a given PFTS (standard approach for CARBONES).
• The Single-Site inversion (SS) results in red as a test to highlight the potential of the model to fit each site separately.
• The prior simulation, using the default ORCHIDEE parameters, in green.
• Top plots represent the normalized time series (satellite NDVI and modelled FPAR scaled between 0 and 1) that are used in the optimization procedure.
• Bottom plots represent the raw time series.
• A little box is inserted in each plot to show the location of the site.
• The Root Mean Square Errors (RMSE) are shown for all sites.
• The prior and posterior values of the four parameters that were optimized in this step are indicated.

Note that only years 2003 to 2006, inclusive, are shown in order to better visualize the differences between curves. One or two sites have been chosen per PFT to demonstrate general patterns seen in the optimisation results.

The results are shown for only a subset of the sites to illustrate the performance of the optimization.

Temperate broad-leaved summergreen:

Boreal broad-leaved summergreen:

Boreal needleleaf summergreen:

C3 grass:

Tropical broad-leaved raingreen:

Go to Top


2. Fit to FluxNet data: Net CO2 and water exchanges

This section illustrates the results of step 2 of the “stepwise optimization” approach using eddy-covariance flux measurements at specific sites (FluxNet network). Only the daily mean CO2 and water fluxes are assimilated.
All graphics in section 2.2 to 2.6 present:

• The eddy-covariance measurements in black
• The Multiple-Site inversion (MS) results in blue that corresponds to the optimisation of all sites at once for a given PFTS (standard approach for CARBONES).
• The Single-Site inversion (SS) results in orange as a test to highlight the potential of the model to fit each site separately.
• The prior simulation, using the default ORCHIDEE parameters, in grey.

Only few sites per PFTs are display to show the strength and weaknesses of the ORCHIDEE model performances.

Go to Top


2.1 Location of the sites

The figure and table below display the location of the different FluxNet sites that were used in the optimization procedure.


Figure 1: Localisation of the different FluxNet sites used within the Carbon Cycle Multi-Data Assimilation System (CCMDAS). Daily net CO2 and water exchange are assimilated.


Table 3: List of FLUXNET sites of Eddy covariance measurements used in Carbones. PFT is the assigned plant functional type in ORCHIDEE. Start and end correspond to the first and last year of the assimilation period. IGBP class is the IGBP land cover for the site and vegetation type is the vegetation or species information provided by the site investigator.

Name Latitude Longitude PFT A start A end IGBP class Vegetation type
AU-Tum -35.655701 148.151993 5 2001 2003 EBF Wet temperate sclerophyl
AU-Wac -37.429001 145.186996 5 2006 2006 EBF Tall cool evergreen forest (Eucalyptus regnans)
BR-Ban -9.824420 -50.159100 2 2004 2005 EBF  
BR-Cax -1.719720 -51.459000 2 2000 2002 EBF Undisturbed forest
BR-Ji2 -10.083200 -61.930901 2 2000 2002 EBF Tropical broad forest
BR-Sa3 -3.018030 -54.971401 2 2001 2002 EBF Cleared forest
BW-Ma1 -19.915501 23.560499 10 1999 2000 SAV Deciduous savannah woodland
CA-Ca3 40.032902 -105.545998 4 2002 2003 ENF Young Douglas fir plantation
CA-Let 49.709301 -112.940002 10 1999 2005 GRA Short/mixed grass prairie (C3/C4)
CA-Man 55.879601 -98.480797 7 1997 2003 ENF Evergreen coniferous forests, boreal needle forest, old black spruce trees
CA-NS1 55.879200 -98.483902 7 2003 2004 ENF Overstory: Closed canopy with 17–20m tall black spruces (Picea mariana [Mill.] BSP); Understory: Open with small numbers of bog Labrador tea (Ledum groenlandicum), Wild rose (Rosa spp.), and Mountain alders (Alnus crispa (Aiton) Pursh), ~70% Deciduous by basal area;
CA-NS2 55.905800 -98.524696 7 2002 2004 ENF Overstory: Closed canopy with 12–20m tall black spruces (Picea mariana Mill.) with a few senescent jack pine (Pinus banksiana Lamb.) and trembling aspen (Populus tremuloides Michx), ~6% Deciduous by basal area; Understory: Significant shrub layer of bog Labrador tea (Ledum groenlandicum), mountain alders (Alnus crispa (Aiton) Pursh), and willow (Salix L.), ~7.2% Deciduous by basal area;
CA-NS3 55.911701 -98.382202 7 2002 2004 ENF  
CA-NS6 55.916698 -98.964401 10 2002 2004 OSH  
CA-NS7 56.635799 -99.948303 10 2003 2004 OSH  
CA-Oas 53.628899 -106.197998 8 2001 2004 DBF Boreal aspen forest
CA-Obs 53.987202 -105.117996 7 2000 2005 ENF Mixed boreal forest, old Jack Pine
CA-Ojp 53.916302 -104.692001 7 2000 2005 ENF Mixed boreal forest, old Jack Pine
CA-Qfo 49.692501 -74.342102 7 2004 2006 ENF Predominantly conifer overstory (black spruce with some jack pine), moss and ericaceous understory)
CA-SJ3 53.875801 -104.644997 7 2005 2005 ENF  
CA-TP4 42.709801 -80.357399 4 2004 2004 ENF Temperate White Pine Forest
CN-HaM 37.369999 101.180000 10 2002 2003 GRA  
CN-Ku1 40.538300 108.694000 5 2006 2006 EBF  
CZ-BK1 49.502602 18.538401 4 2004 2005 ENF  
DE-Bay 50.141899 11.866900 4 1998 1999 ENF Norway spruce (Picea abies), 25 m high, age 55 years (2008); Understory: Calamagrostis villosa, Deschampsia lexuosa, Vaccinium myrtikkus, Dryopteris ilatata, Oxalis acetosella, Dicranum coparium
DE-Geb 51.100101 10.914300 12 2004 2006 CRO  
DE-Hai 51.079300 10.452000 6 2000 2006 DBF Mixed deciduous beech forest
DE-Kli 50.892899 13.522500 12 2005 2006 CRO  
DE-Meh 51.275299 10.655500 10 2004 2005 GRA Grasslands
DE-Tha 50.963600 13.566900 4 1997 2003 ENF Evergreen needleleaf forest
DE-Wet 50.453499 11.457500 4 2002 2006 ENF Evergreen needleleaf forest
DK-Sor 55.486900 11.645800 6 2004 2006 DBF Mixed forest
ES-ES2 39.275501 -0.315220 12 2005 2006 CRO Cropland
ES-LMa 39.941502 -5.773360 10 2004 2005 SAV Evergreen broadleaf forest
ES-VDA 42.152199 1.448500 10 2004 2004 GRA Grassland
FI-Hyy 61.847401 24.294800 7 1997 2006 ENF Evergreen needleleaf forest
FI-Kaa 69.140701 27.295000 10 2000 2006 WET  
FI-Sod 67.361900 26.637800 7 2001 2006 ENF Evergreen needleleaf forest
FR-Aur 43.549400 1.107780 12 2005 2005 CRO  
FR-Fon 48.476299 2.780150 6 2006 2006    
FR-Gri 48.844002 1.952430 12 2005 2006 CRO Cropland
FR-Hes 48.674198 7.064620 6 2001

2003

DBF Deciduous broadleaf forest, beech
FR-Lam 43.493301 1.237220 12 2005 2005 CRO  
FR-LBr 44.717098 -0.769300 4 2003 2006 ENF Evergreen needleleaf forest, Mediterranean/montane, evergreen coniferous plantation
HU-Bug 46.691101 19.601299 10 2003 2006 GRA Grassland
HU-Mat 47.846901 19.726000 10 2004 2006 GRA Cropland
ID-Pag 2.345000 114.036003 2 2002 2003 EBF  
IE-Ca1 52.858799 -6.918140 12 2004 2006 CRO  
IE-Dri 51.986698 -8.751810 10 2003 2004 GRA Grassland
IT-Amp 41.904099 13.605200 10 2005 2005 GRA Grassland
IT-BCi 40.523800 14.957400 12 2005 2006 CRO  
IT-Lav 45.955299 11.281200 4 2004 2004 ENF Mixed forest
IT-Mal 46.116699 11.702800 10 2003 2004 GRA  
IT-MBo 46.015598 11.046700 10 2004 2006 GRA  
IT-Ren 46.587799 11.434700 4 2002 2002 ENF Evergreen needleleaf forest
IT-SRo 43.727859 10.284440 4 2002 2004 ENF  
JP-Mas 36.053970 140.026917 12 2002 2003 CRO  
JP-Tak 36.146198 137.423004 6 1999 2004 DBF  
KR-Hnm 34.549999 126.570000 12 2006 2006 CRO  
NL-Ca1 51.971001 4.927000 10 2003 2004 GRA  
NL-Hor 52.028900 5.067500 10 2004 2006 GRA  
NL-Lan 51.953602 4.902900 12 2005 2005 CRO  
NL-Loo 52.167900 5.743960 4 2001 2002 ENF Evergreen coniferous forests, spruce, evergreen
PL-wet 52.762199 16.309401 10 2004 2005 WET Wetland
SE-Abi 68.362389 18.794750 8 2005 2005 DBF  
SE-Deg 64.183296 19.549999 10 2001 2005 WET  
SE-Fla 64.112801 19.456900 7 2001 2002 ENF Evergreen coniferous forests
SE-Nor 60.086498 17.479504 4 1996 1997 ENF Evergreen coniferous forests
SE-Sk1 60.125000 17.918100 4 2005 2005 ENF Evergreen needleleaf forest
SE-Sk2 60.129669 17.840059 4 2005 2005 ENF  
UK-ESa 55.906940 -2.858610 12 2005 2005 CRO  
UK-Gri 56.607220 -3.798060 4 2000 2001 ENF Intensively managed plantation, coniferous forest
UK-Ham 51.120800 -0.860830 6 2004 2005 DBF Deciduous broadleaf forest
US-ARM 36.605000 -97.488403 10 2003 2005 CRO  
US-Aud 31.590700 -110.510002 10 2005 2005 GRA  
US-Bar 44.064602 -71.288078 6 2004 2005 DBF  
US-Bkg 44.345299 -96.836197 10 2005 2006 GRA  
US-Bn1 63.919800 -145.378006 7 2003 2003 ENF Black spruce (Picea mariana)
US-Bn2 63.919800 -145.378006 8 2003 2003 DBF Deciduous broadleaf forest, burn site
US-Bn3 63.922699 -145.744003 10 2003 2003 OSH Black spruce (Picea mariana) and bunch grasses (Festuca altaica)
US-Goo 34.250000 -89.970001 10 2004 2004 GRA Dominantly short grasses with scattered trees and shrubs
US-Ha1 42.537800 -72.171501 6 2003 2006 DBF  
US-Ho1 45.204102 -68.740303 4 2003 2004 ENF The natural stands in this boreal--northern hardwood transitional forest consist of hemlock-spruce-fir, aspen-birch, and hemlock-hardwood mixtures.
US-Ho2 45.209099 -68.747002 4 1999 2004 ENF  
US-IB1 41.859299 -88.222702 13 2006 2006 CRO Corn/Soybean rotation
US-IB2 41.840599 -88.240997 10 2006 2006 GRA  
US-Ivo 68.486504 -155.750000 10 2004 2005 OSH  
US-LPH 42.541901 -72.184998 6 2003 2004 DBF  
US-Me2 44.452400 -121.556999 4 2004 2005 ENF Dominantly ponderosa pine with scarce incense cedar
US-Me4 44.499199 -121.622002 4 2000 2000 ENF Overstory is composed of 100% ponderosa pine
US-MOz 38.744099 -92.199997 6 2005 2006 DBF Oak hickory forest
US-NC2 35.803101 -76.667900 4 2005 2006 ENF  
US-Ne1 41.165100 -96.476601 13 2002 2004 CRO Agriculture (continuous maize)
US-Ne2 41.164902 -96.470100 13 2003 2004 CRO Agriculture (maize-soybean rotation)
US-Ne3 41.179699 -96.439697 13 2004 2004 CRO Agriculture (maize-soybean rotation)
US-NR1 40.032902 -105.545998 7 2002 2003 ENF Subalpine mixed coniferous forest with very little understory
US-SRM 31.821400 -110.865997 10 2004 2006 WSA 30% canopy coverage by Velvet mesquites, typically 3-4 m high. The majority of understory is dominated by a series of desert and savanna grasses. Maximum LAI measurements occur during mid-monsoon season, JD 213-243, for all understory and overstory species.
US-UMB 45.559799 -84.713799 6 2000 2003 DBF  
US-Var 38.413300 -120.950729 10 2002 2002 GRA  
US-WCr 45.805901 -90.079903 6 1999 2004 DBF Mature sugar maple-aspen-yellow birch forests
US-Wrc 45.820499 -121.952003 4 1999 2002 ENF Douglas-fir and western hemlock transitional overstory lies between the Western Hemlock Zone and the Pacific Silver Fir Zone, with an understory dominated by vine maple, salal, and Oregon grape.

 

Go to Top


2.2 Temperate broadleaf deciduous forests

DE-Hai:

FR-Fon:

FR-HES:

JP-Tak:

US-Bar:

Go to Top


2.3 Temperate broadleaf evergreen forests

AU-Tum:

AU-Wac:

Go to Top


2.4 Boreal coniferous evergreen forests

CA-Man:

CA-NS2:

CA-Ojp:

FI-Hyy:

SE-Fla:

Go to Top


2.5 Boreal broadleaf deciduous forests

CA-Oas:

SE-Abi:

Go to Top


2.6 Herbaceous C3 grassland

CA-Let:

CN-Ham:

ES-LMa:

HU-Mat:

FI-Kaa:

IT-Mal:

US-Bkg:

Go to Top


2.7 Tropical evergreen forests

BR-Cax:

BR-Ji2:

BR-Sa3:

Go to Top


3. Fit to ocean surface pCO2 data

This section corresponds to the ocean component of the CCDAS that computes air-sea CO2 fluxes by using ocean surface partial pressure of CO2 (pCO2oce) as driving observations. Modeled pCO2oce are computed with an artificial neural network (referred as OCVR) that relates pCO2 to explanatory variables (Sea surface temperature, Sea surface Salinity, Mix layer Depth, Chlorophyll content observed from space). A first calibration step consist to adjust the seasonal component of the pCO2 by fitting the model to the reference climatological data of Takahashi et al. (2009) for the year 2000. In a second step, the interannual variability of pCO2 as well as singular events (e.g., El Nino) are introduced through an implicit 4D-var scheme to efficiently incorporate from 1989 up to the 2009 about 4 millions of raw pCO2 data (Takahashi et al. 2009), previously gridded at monthly and 2°x2° spatial resolutions.

The figure below displays the residuals between the simulated and modelled pCO2 (after the overall calibration) obtained on the validation subset of the database: 25% of the data were left over (not use in the calibration of the neural network) and kept for evaluation. The red curve represents the residuals in micro-atmosphere from OCVR model, and the blue curve the same residuals but from the standard Takahashi 2009 climatology extrapolated in time with a fixed trend.


Figure 2: Performance on the validation database (25% randomly excluded from the raw database): error differences expressed in µatm along the reanalysis period for i) the climatology of Takahashi et al. 2009 after linear extrapolation in time in blue (Mean = -0.56, Std = 18.15) and ii) the results of OCVR in red (Mean = -0.06, Std = 16.41).

Go to Top


4. Fit to the atmospheric CO2 concentrations

We display below the results of the model data fit to the atmospheric CO2 observations that corresponds to step 3 of the “stepwise optimization approach”. The fits presented below correspond to the specific case where the fluxes themselves are directly optimized in step 3 and not the parameters of the model.

Go to Top


4.1 Location of the surface stations

The map below displays the location of the different surface stations that are used in CARBONES CCDAS.


Figure 3: Localization of the different stations used for the first version of the CCDAS

Go to Top


4.2 Fit to the atmospheric CO2 concentrations

The following figures display in black the observed concentrations, in green the prior simulated concentrations, and in red the optimized concentrations. The vertical bars in grey correspond to the error bars associated to the observations.


Go to Top