Using a Global Flux Network—FLUXNET— to Study the Breathing of the Terrestrial Biosphere

Dennis BaldocchiUniversity of California, Berkeley

BrasFlux, Santa Maria Rio Grande do Sul, Brazil

Nov. 14, 2011

Contemporary CO2 Record

Mauna LoaKeeling data

year

1950 1960 1970 1980 1990 2000 2010

CO

2 (pp

m)

300

310

320

330

340

350

360

370

380

Methods To Assess Terrestrial Carbon Fluxes at Landscape to Continental Scales, and Across Multiple

Time Scales

GCM InversionModeling

Remote Sensing/MODIS

Eddy Flux Measurements/FLUXNET

Forest/Biomass Inventories

Biogeochemical/Ecosystem Dynamics Modeling

Physiological Measurements/Manipulation Expts.

remote sensingof CO2

Tem

pora

l sca

le

Spatial scale [km]

hour

day

week

month

year

decade

century

local 0.1 1 10 100 1000 10 000 globalContinentplot/site

talltowerobser-

vatories

Eddycovariance

towers

Landsurface remote sensing

Developing a System that is Everywhere, All of the Time

Credit: Markus Reichstein, MPI

Status of Global Network, 500+ Sites

FLUXNET

FLUXNET Represents Many Climate Spaces, Well

More Data and Sites in Tropical Rain Forests are Welcome!

1980s 1995s 2000s 2010s

30 hours

One Site-Year

1990s

Ten Site-Years

100 Site-Years

1000 Site-Years

Time Line of Flux Data

Growth in Africa, Australia and Asia, Sustenance in Brazil and EuropeDecline in Canada and USVoids in India and Latin America

Many Towers are Not Active, nor Submitting data, circa La Thuile dataset

FLUXNET Data Archive

Year

1990 1992 1994 1996 1998 2000 2002 2004 2006 2008

Site

-Yea

rs o

f Dat

a

0

20

40

60

80

100

120

140

160

180

200

Fluxdata.org – A Common, Shared Database

Individual Sites

Regional Networks

Flux/MetData

www.Fluxdata.org

BADMData

Flux/MetData

BADM Data + Flux/MetGap-Filled/QA/Products

Data processing, Value Added Products and Uncertainty Estimation

Half hourly data

u* threshold selection- 3/5 different methods

u* filtering - 3/4 possibility

Gapfilling- 2 methods

Partitioning- 3 methods

3 methods, bootstraping…

Also daytime, also data after low turb., …

MDS and ANN

Reichstein, Lasslop, van Gorsel

Community Building/OutReach

Shared Database: www.fluxdata.org

Newsletter: FluxLetter

Young Scientist Forums

International WorkshopsMarconi, 2000Orvieto, 2002Lake Tahoe, 2003Firenze, 2004La Thuile, 2007Asilomar, 2009Berkeley, 2011

Challenges/Opportunities for Future

• To Sustain and Grow the Network that Asks and Answers Network-Scale Questions– To sample representative Climates and Biomes– To sample representative Disturbance Classes– Detect trends in Fluxes as Climate and Land Use Changes– Validate and Parameterize New Generation of Land Surface-

Atmosphere Exchange Models– Serve as Critical Partner in Machine Learning Approaches to

Flux Upscaling with Satellite Remote Sensing– Provide Ground Truth for MODIS Land Products– Provide Ground Truth for Sources/Sinks produced by global

CO2 networks and next generation CO2 Satellite Sensors

Big Network Issues Facing Flux Networks in 2011

• How can we Sustain the Network?• How many Stations are Enough?• Where do we need new Stations?• How can we encourage Scientists to

Participate, Submit Data and Open the Data to the Wider Science Community, with Benefits to ALL?

Data Sharing is a Win/Win Activity

• Get Access to Data from other Sites and Networks• Learn more about your site though Comparisons• Can Initiate, Lead or Participate on multi-site

synthesis activities• Data Sharing Rules Come from the Community via

Bottom-Up Process– Multiple Levels of Data Availability are Provided– Proposed Projects are Vetted to Protect PIs and Reduce

Duplication• Future of Biogeosciences and Flux Research

Charge/Recommendations for BrasFlux

• Roving Network Calibration System• Expand Suite of Sensors

– Digital Camera for Phenology and LAI– Diffuse Radiation– 4 Band Net Radiometers and High Quality

Quantum Sensor• Database, with Vetted Processing

Algorithms, Qa/Qc, Comprehensive Site Documentation and Version Control

What Have We Learned?

• Time– Annual Integration– Seasonal Dynamics– Inter-Annual Variability– Disturbance/Chronosequence

• Processes– Photosynthesis = f(Q,T,functional type)– Respiration = f(T, growth, ppt, q)

• Space• Other Uses and Application

– Ecosystem Modeling

Published Data, 2011

NEE (gC m-2 y-1)

-1000 -500 0 500 1000

pdf

0.000

0.002

0.004

0.006

0.008

0.010

0.012

0.014

0.016

n = 973mean = -165 +/- 253 gC m-2 y-1

Probability Distribution of Published NEE Measurements, Integrated Annually

Baldocchi, Austral J Botany, 2008

Does Net Ecosystem Carbon Exchange Scale with Photosynthesis?

FA (gC m-2 y-1)

0 500 1000 1500 2000 2500 3000 3500 4000

F N (g

C m

-2 y

-1)

-1000

-750

-500

-250

0

250

500

750

1000

Ecosystems with greatest GPP don’t necessarily experience greatest NEE

FA (gC m-2 y-1)

0 500 1000 1500 2000 2500 3000 3500 4000

F R (g

C m

-2 y

-1)

0

500

1000

1500

2000

2500

3000

3500

4000

UndisturbedDisturbed by Logging, Fire, Drainage, Mowing

Baldocchi, Austral J Botany, 2008

Ecosystem Respiration Scales Tightly with Ecosystem Photosynthesis, But Is with Offset by Disturbance

FLUXNET 2007 Database

GPP at 2% efficiency and 365 day Growing Season

Are Large Carbon Fluxes Defensible?

tropics

GPP at 2% efficiency and 182.5 day Growing Season

Length of Growing Season, days

50 100 150 200 250 300 350

F N (g

C m

-2 y

r-1)

-1000

-800

-600

-400

-200

0

200

Temperate and Boreal Deciduous Forests Deciduous and Evergreen Savanna

Baldocchi, Austral J Botany, 2008

Net Ecosystem Carbon Exchange Scales with Length of Growing Season

Harvard Forest, 1991-2004

Year

1990 1992 1994 1996 1998 2000 2002 2004 2006

NE

E (g

C m

-2 d

-1)

-10

-8

-6

-4

-2

0

2

4

6

8

10

Data of Wofsy, Munger, Goulden, et al.

Decadal Plus Time Series of NEE:Flux version of the Keeling’s Mauna Loa Graph

Interannual Variation and Long Term Trends in Net Ecosystem Carbon Exchange (FN), Photosynthesis (FA) and Respiration (FR)

Urbanski et al 2007 JGR

Harvard Forest

Year

1990 1992 1994 1996 1998 2000 2002 2004 2006 2008

Car

bon

Flux

Den

sity

, gC

m-2

y-1

-600

-400

-200

0

800

1000

1200

1400

1600

1800

GPPNEEReco

Interannual Variability in FN

d FA/dt (gC m-2 y-2)

-750 -500 -250 0 250 500 750 1000

d F R

/dt (

gC m

-2 y

-2)

-750

-500

-250

0

250

500

750

1000Coefficients:b[0] -4.496b[1] 0.704r ² 0.607n =164

Baldocchi, Austral J Botany, 2008

Interannual Variations in Photosynthesis and Respiration are Coupled

Perturbations in Fluxes following 2003 European Heat Spell/Drought

Implications on Drying of the Amazon

How many Towers are needed to estimate mean NEE, GPPand assess Interannual Variability, at the Global Scale?

We Need about 75 towers to produce Robust and Invariant Statistics Based on Current Population

Increasing the Size of the Network Reduces the Sampling Error, but in an Asymptotic Manner

Limit in the Precision of NEE Change that can be detected if Upscaled Globally:+/- 20 gC m-2 y-1 ~ 2 PgC/y = 2 1015 gC/y

Can We Truly Detect Year-Year Variations in Fluxes with a Sparse Network?

Errors that sound Small at one scale may be Huge at another..

Says Nothing about Biases by Under Sampling Dominant Regions like the Tropics

Interannual Variability in NEE is tiny across the Global Network

FLUXNET Network, 75 sites

NEE (gC m-2 y-1)

-1000 -800 -600 -400 -200 0 200 400 600

p(N

EE

)

0.00

0.05

0.10

0.15

0.20

0.25

2002: -220 +/- 35.2 gC m-2 y-1

2003: -238 +/- 39.92004: -243 +/- 39.7 2005: -237 +/- 38.7

Year

1996 1998 2000 2002 2004 2006

C F

lux,

gC

m-2

y-1

-500

-400

-300

-200

-100

0

Ne

What is Interannual Variability of Fluxes, sampled with the Network and the Network Detection Limit?

This Analysis Would Suggest Global Metabolism is Invariant with Time, like the Solar Constant

FLUXNET database, means +/- 95% C.I.

Year

1996 1998 2000 2002 2004 2006

C F

lux,

gC

m-2

y-1

400

600

800

1000

1200

1400

GPPReco

Assuming Global Arable Land area is 110 106 km2, Mean Global GPP ranges between 121.3 and 127.8 PgC/y

Precision is about +/- 7 PgC/y

Complicating Dynamical Factors

• Switches– Phenology– Drought– Frost/Freeze

• Pulses– Rain– Litterfall

• Emergent Processes– Diffuse Light/LUE

• Acclimation• Lags • Stand Age/Disturbance

Temperate Broadleaved Deciduous Forest

Day

0 50 100 150 200 250 300 350

NE

E (g

C m

-2 d

-1)

-7

-6

-5

-4

-3

-2

-1

0

1

2

3

4

5

LAI=0GPP=0;Litterfall (+)Reco=f(litterfall)(+)

snow:Tsoil(+)

GPP=0; Reco(+)

no snow

Tsoil (-)

Reco (-)

GPP=f(LAI, Vcm

ax )

late spring

early spring

Drought:q(-)GPP(-); Re(-)

Clouds:PAR(-) GPP=f(PAR)(+)

Emergent Scale Process:CO2 Flux and Diffuse Radiation

Niyogi et al., GRL 2004

• We are poised to see effects of Cleaner/Dirtier Skies and Next Volcano

Mean Summer Temperature (C)5 10 15 20 25 30

Tem

pera

ture

Opt

imum

fo

r Can

opy

CO 2 upt

ake

(C)

5

10

15

20

25

30

35

b[0] 3.192b[1] 0.923r ² 0.830

E. Falge et al 2002 AgForMet; Baldocchi et al 2001 BAMS

Optimal NEE: Acclimation with Temperature

Soroe, DenmarkBeech Forest1997

day

0 50 100 150 200 250 300 350-10

-5

0

5

10

15

20

NEE, gC m-2 d-1

Tair, recursive filter, oC

Tsoil, oC

Data of Pilegaard et al.

Soil Temperature: An Objective Indicator of Phenology??

Baldocchi et al. Int J. Biomet, 2005

Soil Temperature: An Objective Measure of Phenology, part 2

Temperate Deciduous Forests

Day, Tsoil >Tair

70 80 90 100 110 120 130 140 150 160

Day

NEE=

0

70

80

90

100

110

120

130

140

150

160

DenmarkTennesseeIndianaMichiganOntarioCaliforniaFranceMassachusettsGermanyItalyJapan

Day after rain (d)-5 0 5 10 15 20

Rec

o (gC

m-2

d-1)

0

2

4

6

8

10DOY311 2002 understoryDOY214 2003 understoryDOY311 2002 grasslandDOY214 2003 grassland

A

Pre-rain Reco (gC m-2d-1)0.0 0.5 1.0 1.5 2.0 2.5

Rec

o enh

ance

meb

t (gC

m-2

d-1)

0

2

4

6

8B

Rain-Induced Respiration Pulses

Xu et al. 2005 GBC

Spatial Variations in C Fluxes

Xiao et al. 2008, AgForMet

) )

mmk

Pk

mmMATaa

Caa

C eeGPP

eeGPP

PgMATfGPP

10001000

15

15 11,

11min

,min

21

21

Upscale NEP, Globally, Explicitly

1. Compute GPP = f(T, ppt)2. Compute Reco = f(GPP,

Disturbance)3. Compute NEP = GPP-Reco

Leith-Reichstein Model

Reco = 101 + 0.7468 * GPP

Reco, disturbed= 434.99 + 0.922 * GPPFA (gC m-2 y-1)

0 500 1000 1500 2000 2500 3000 3500 4000

F R (g

C m

-2 y

-1)

0

500

1000

1500

2000

2500

3000

3500

4000

UndisturbedDisturbed by Logging, Fire, Drainage, Mowing FLUXNET Synthesis

Baldocchi, 2008, Aust J Botany

FLUXNET Database

NEE (gC m-2 y-1)

-1400 -1200 -1000 -800 -600 -400 -200 0 200 400 600

pdf

0.00

0.02

0.04

0.06

0.08

0.10

0.12

0.14

0.16

0.18

0.20

FLUXNETGlobal Map

NEE, FLUXNET = -225 gC m-2 y-1

NEE, globally-integrated, area-wt = -129 gC m-2 y-1

Pros and Cons of Extracting Global Information from a Sparse Network

FLUXNET Database

GPP (gC m-2 y-1)

0 1000 2000 3000 4000

pdf

0.00

0.02

0.04

0.06

0.08

0.10

0.12

0.14

FLUXNET (254 sites-years), <GPP> = 1033 gC m-2 y-1 Cartesian-Gridded, <GPP> =1139 g C m-2 y-1

Area-Weighted, <GPP> = 1281 gC m-2 y-1

FLUXNET Over represents GPP in Temperate, Mid-Productive Ecosystems; Under-represents GPP in Semi-Arid, low-productive and

Tropical, High Productive Regions

8 day means

Daily g

ross

CO

2 flux

(m

mol m

-2 d

ay-1

)

0

200

400

600

800

1000

1200

1400

16008 day meansDaily n

et C

O2 flux

(m

mol m

-2 d

ay-1

)

-400

-200

0

200

400

600

800

r2 = 0.92

8 day means

Daily g

ross

LUE

0.00

0.01

0.02

0.03

r2 = 0.65

Single clear days

AM net CO2 flux (mmol m-2 hr-1)

0 20 40 60 80 100 120

Daily n

et C

O2 flux

(m

mol m

-2 d

ay-1

)

-400

-200

0

200

400

600

800Single clear days

AM gross CO2 flux (mmol m-2 hr-1)

0 20 40 60 80 100 120 140

Daily g

ross

CO

2 flux

(m

mol m

-2 d

ay-1

)

0

200

400

600

800

1000

1200

1400

1600

r2 = 0.88

Single clear days

AM gross LUE

0.00 0.01 0.02 0.03Daily g

ross

LUE

0.00

0.01

0.02

0.03

r2 = 0.73

r2 = 0.64

r2 = 0.56

Evergreen needleleaf forestDeciduous broadleaf forestGrassland and woody savanna

a b c

d e f

Sims et al 2005 AgForMet

Do Snap-Shot C Fluxes, inferred from Remote Sensing, Relate to Daily C Flux Integrals?

UpScaling of FluxNetworks

Beer et al. 2010 Science

Global GPP121 +/- 8 PgC/y

0

500

1000

1500

2000

2500

3000

12

34

56

78

-10-505101520

GP

P (g

C m

-2 y

-1)

Rg (G

J m-2 y

-1 )

Tair (C)

FLUXNET Database

0 500 1000 1500 2000 2500 3000

Joint pdf GPP, Solar Radiation and Temperature

E[GPP]= 1237 gC m-2 y-1~136 PgC/y

Importance, and Uncertainty, of Tropical GPP

Beer et al. 2010 Science

Regional Maps for Carbon Markets

Disturbances

Climate Anomalies

Jung et al. 2010 Nature

Evaluation of upscaled global Evapotranspiration

(a) Map of mean Evapotranspiration from 1982-2008

(b) Predicted vs. Observed ET at FLUXNET sites (10-fold cross-validation from MTE training)

(c) Corroboration aganist river catchment water balances

(d) Comparison against GSWP-2 land surface model ensemble (16 models) stratified according to bioclimatic zones

Baldocchi, White, Schwartz, unpublished

Spatialize Phenology with Transformation Using Climate Map

Mean Air Temperature, C

4 6 8 10 12 14 16 18

Day

of NEE

= 0

60

80

100

120

140

160

Coefficients:b[0]: 169.3b[1]: -4.84r ²: 0.691

Limits to Landscape Classification by Functional Type

• Stand Age/Disturbance• Biodiversity• Fire• Logging• Insects/Pathogens• Management/Plantations• Kyoto Forests

Conifer Forests, Canada and Pacific Northwest

Stand Age After Disturbance

1 10 100 1000

F N (g

C m

-2 y

-1)

-600

-400

-200

0

200

400

600

800

1000

Time Since Disturbance Affects Net Ecosystem Carbon Exchange:What Happens in the Tropics?

Baldocchi, Austral J Botany, 2008 Data of teams lead by Amiro, Dunn, Paw U, Goulden

Other Activities and Uses of Fluxnet Data

• Landuse• Ecosystem Modeling• EcoHydrology• Biodiversity• Climate Modeling

Biodiversity and Evaporation

Temperate/Boreal Broadleaved ForestsSummer Growing Season

Number of Dominant Tree Species (> 5% of area or biomass survey)

1 2 3 4 5 6 7 8

E/Eeq

0.5

0.6

0.7

0.8

0.9

1.0

1.1

1.2

1.3

Baldocchi, 2004: Data from Black, Schmid, Wofsy, Baldocchi, Fuentes

Biodiversity and Evaporation on Annual Time Scales

Number of Species

0 1 2 3 4 5 6 7 8

LE/R

n

0.0

0.2

0.4

0.6

0.8

1.0

1.2

Coefficients:b[0]0.5918072042b[1]-0.0316243545r ²0.1263967329

Deciduous Temperate Forest

Baldocchi, unpublished

Kucharik et al., 2006 Ecol Modeling

Ecosystem Model Testing and Development

Seasonality of Photosynthetic Capacity

Wang et al, 2007 GCB

Optimizing Seasonality of Vcmax improves Prediction of Fluxes

Wang et al, 2007 GCB

Bonan et al 2011, JGR Biogeoscience

Improvement in CLM-4

Via better radiative transfer modeling, VCmx and stomatal Conduct

Model structural revisions reduce global GPPover the period 1982–2004 from 165 Pg C yr−1 to 130 Pg C yr−1, and globalevapotranspiration decreases from 68,000 km3 yr−1 to 65,000 km3 yr−1,

Improvements in CLM vs Fluxnet upscaling

Important to Sustain Networks Because Big Science Questions Being Answered by Flux Networks

• What is Global GPP?– 123 +/- 8 PgC/ y, Beer et al, Science

• What is Global ET?– 65,000 km3/y, Jung et al. 2010, Nature

• What is Global Year to Year Variability in GPP and NEE?• What Emergent Scale Properties arise at the Ecosystem Scale?

– Rain-induced respiration Pulses, modulated by photodegradation

– Diffuse Light Enhances LUE– Q10 are static and Respiration Decrease with Heat and

Drought– Optimal temperature of Photosynthesis Acclimates with local

Climate– What is the effect of time since disturbance on NEE of

tropical regions?

Acknowledgements

• Founding Leadership– Riccardo Valentini, Steve Running

• Data Preparation: FLUXNET-2007– Dario Papale, Markus Reichstein, Catharine Van Ingen,

Deb Agarwal, Tom Boden, Bob Cook, +++• FLUXNET Office @ Berkeley

– Eva Falge, Lianhong Gu, Matthias Falk, Rodrigo Vargas, Laurie Koteen

• Regional Networks – AmeriFlux, CarboEurope, AsiaFlux, ChinaFlux, Fluxnet

Canada, OzFlux, LBA, +++• Agencies

– NSF/RCN, ILEAPS, DOE/TCP, NASA, Microsoft, ++++