estimates of carbon transfer coefficients using probabilistic inversion for three forest ecosystems...

1
Estimates of Carbon Transfer coefficients Using P robabilistic Inversion for Three Forest Ecosystem s in East China Li Zhang 1 , Yiqi Luo 2 , Guirui Yu 1 , Leiming Zhang 1 1 Key Laboratory of Ecosystem Network Observation and Modeling, Institute of Geographic Science and Natural Resources Research, Chinese Academy Sciences 2 Department of Botany and Microbiology, University of Oklahoma, Norman, USA Carbon transfer coefficients are the key parameters in carbon cycling models for quantifying the capacity of C leaving in each pool at a time step. They can be used to estimate the residence time of carbon, which determines the capacity of an ecosystem as a carbon source or sink. Here we will apply MCMC parameter estimation technique to inverse carbon transfer coefficients for eight pools against with measurements of carbon pools and carbon fluxes at three forest sites, and compare the estimates of parameters constrained by different sets of assimilation data. Introduction Discussions Eddy covariance net ecosystem exchange of CO 2 (NEE) can provide with useful information for constraining carbon transfer rates between litter, microbes, slow SOM and passive SOM, but with little information for constraining carbon transfer rates from leaf, fine root and woody biomass to litter pool. The effect of NEE data on parameters c4-c8 may result from its lower weight in the cost function compared with other observed data. Methods C anopy photosynthesis Foliage Biom ass (X 1 ) Fine R ootBiom ass (X 2 ) W oody Biom ass (X 3 ) M etabolic Litter (X 4 ) Structure Litter (X 5 ) M icrobes (X 6 ) Slow SO M (X 7 ) Passive SO M (X 8 ) co 2 co 2 co 2 co 2 co 2 co 2 co 2 co 2 C anopy photosynthesis Foliage Biom ass (X 1 ) Fine R ootBiom ass (X 2 ) W oody Biom ass (X 3 ) M etabolic Litter (X 4 ) Structure Litter (X 5 ) M icrobes (X 6 ) Slow SO M (X 7 ) Passive SO M (X 8 ) co 2 co 2 co 2 co 2 co 2 co 2 co 2 co 2 8-pool modified TECOS model Table 1 Description of Carbon Transfer Coefficients Paramet er Description c1 from pool “nonwoody biomass” to “metabolic litter” and “structure litter” c2 from pool “fine root biomass” to “metabolic litter” and “structure litter” c3 from pool “woody biomass” to “structure litter” c4 from pool “metabolic litter” to “microbes” c5 from pool “structure litter” to “microbes” and “slow SOM” c6 from pool “microbes” to “slow SOM” and “passive SOM” c7 from pool “slow SOM” to “microbes” and “passive SOM” c8 from pool “passive SOM” to “microbes” Data sets The data sets used here are biomass (foliage, fine root, woody) , litterfall, soil organic C, soil respiration, n et ecosystem exchange of CO 2 (NEE) at CBS, QYZ and DHS si tes. Parameter estimation Markov Chain Monte Carlo (MCMC) method was used to estimate the carbon transfer coefficients (Table 1). Three experiments were undertaken with different sets of assimilation data . OBS1 : Assimilating carbon pools, soil respiration and NEE (all data) OBS2 : Assimilating carbon pools and soil respiration OBS3: Assimilation NEE only Study sites CBS QYZ DHS Table 2 Site location and long-term climate variables Code CBS QYZ DHS Location 42°24′N, 128°06′E 26°44′N, 115°04′E 23°10′N, 112°32′E Terrain flat hill upland Elevation (m) 736 100 300 Annual Precipitation (mm) 695 1485 1956 Average temperature (℃) 4.0 17.9 21.1 Canopy height (m) 26 11 17 Age (yr) 200 24 100 CBS: a broad-leaved and Korean pine mixed forest. QYZ: a young evergreen coniferous plantation. DHS: an evergreen conifer and broad- leaved mixed forest. Results 0 10 20 30 40 50 Frequency (% ) O bs1 0 10 20 30 40 50 O bs2 2 4 6 0 10 20 30 40 50 C1(× 10 -3 ) O bs3 1 2 C2(× 10 3 ) 1 2 C3(× 10 4 ) 1 2 C4(× 10 2 ) 1 2 C5(× 10 3 ) 3 4 5 6 C6(× 10 3 ) 1 2 C7(× 10 4 ) 2 4 6 8 C8(× 10 6 ) 0 10 20 30 40 O bs2 0 10 20 30 40 Frequency(% ) O bs1 1 2 0 10 20 30 40 C1(× 10 3 ) O bs3 1 2 C2(× 10 3 ) 1 2 C3(× 10 4 ) 1 2 C4(× 10 2 ) 1 2 C5(× 10 3 ) 3 4 5 6 C6(× 10 4 ) 1 2 C7(× 10 4 ) 2 4 6 8 C8(× 10 6 ) 0 10 20 30 O bs2 0 10 20 30 40 50 Frequency(% ) O bs1 1 2 0 10 20 30 40 c1(× 10 3 ) O bs3 1 2 c2(× 10 3 ) 1 2 c3(× 10 4 ) 1 2 c4(× 10 2 ) 1 2 c5(× 10 3 ) 3 4 5 6 c6(× 10 3 ) 1 2 c7(× 10 4 ) 2 4 6 8 c8(× 10 6 ) Our results showed that the estimates of parameters c1, c2 and c3 will not be influenced by NEE data, but constrained by carbon pools data. Because there is no CO 2 release when carbon transfers from leaf, fine root and woody biomass to litter pool. On the contrary, carbon transfers between litter, microbes, slow SOM and passive SOM accompanied with CO2 release. Thus, the estimates of parameters c4, c5, c6, c7 and c8 changed when adding NEE to data sets. Fig. 1 Posterior Distribution of parameters with 3 different assimilation experiments in CBS site Fig. 2 Posterior Distribution of parameters with 3 different assimilation experiments in QYZ site Fig. 3 Posterior Distribution of parameters with 3 different assimilation experiments in DHS site Acknowledgements This work was supported by Chinese Academy of Sciences International P artnership Project "Human Activitie s and Ecosystem Changes" (No CXTD-Z 2005-1). We thank all related staff s of ChinaFLUX and CERN for their contribution from observation to data processing.

Upload: cameron-powers

Post on 13-Dec-2015

213 views

Category:

Documents


0 download

TRANSCRIPT

Page 1: Estimates of Carbon Transfer coefficients Using Probabilistic Inversion for Three Forest Ecosystems in East China Li Zhang 1, Yiqi Luo 2, Guirui Yu 1,

Estimates of Carbon Transfer coefficients Using Probabilistic Inversion for Three Forest Ecosystems in East China

Li Zhang1, Yiqi Luo2, Guirui Yu1, Leiming Zhang1

1 Key Laboratory of Ecosystem Network Observation and Modeling,

Institute of Geographic Science and Natural Resources Research, Chinese Academy Sciences2 Department of Botany and Microbiology, University of Oklahoma, Norman, USA

Carbon transfer coefficients are the key parameters in carbon cycling models for quantifying the capacity of C leaving in each pool at a time step. They can be used to estimate the residence time of carbon, which determines the capacity of an ecosystem as a carbon source or sink. Here we will apply MCMC parameter estimation technique to inverse carbon transfer coefficients for eight pools against with measurements of carbon pools and carbon fluxes at three forest sites, and compare the estimates of parameters constrained by different sets of assimilation data.

Introduction

Discussions

Eddy covariance net ecosystem exchange of CO2 (NEE) can provide with useful information for constraining carbon transfer rates between litter, microbes, slow SOM and passive SOM, but with little information for constraining carbon transfer rates from leaf, fine root and woody biomass to litter pool. The effect of NEE data on parameters c4-c8 may result from its lower weight in the cost function compared with other observed data.

Methods

Canopy photosynthesis

Foliage Biomass(X1)

Fine Root Biomass(X2)

Woody Biomass(X3)

Metabolic Litter(X4)

Structure Litter(X5)

Microbes(X6)

Slow SOM(X7)

Passive SOM(X8)

co2

co2 co2

co2 co2

co2

co2

co2

Canopy photosynthesis

Foliage Biomass(X1)

Fine Root Biomass(X2)

Woody Biomass(X3)

Metabolic Litter(X4)

Structure Litter(X5)

Microbes(X6)

Slow SOM(X7)

Passive SOM(X8)

co2

co2 co2

co2 co2

co2

co2

co2

8-pool modified TECOS model

Table 1 Description of Carbon Transfer Coefficients

Parameter Description

c1 from pool “nonwoody biomass” to “metabolic litter” and “structure litter”

c2 from pool “fine root biomass” to “metabolic litter” and “structure litter”

c3 from pool “woody biomass” to “structure litter”

c4 from pool “metabolic litter” to “microbes”

c5 from pool “structure litter” to “microbes” and “slow SOM”

c6 from pool “microbes” to “slow SOM” and “passive SOM”

c7 from pool “slow SOM” to “microbes” and “passive SOM”

c8 from pool “passive SOM” to “microbes”

Data sets

The data sets used here are biomass (foliage, fine root, woody) , litterfall, soil organic C, soil respiration, net ecosystem exchange of CO2 (NEE) at CBS, QYZ and DHS sites.

Parameter estimationMarkov Chain Monte Carlo (MCMC) method was used to estimate the carbon transfer coefficients (Table 1). Three experiments were undertaken with different sets of assimilation data .OBS1 : Assimilating carbon pools, soil respiration and NEE (all data)OBS2 : Assimilating carbon pools and soil respirationOBS3: Assimilation NEE only

Study sites

CBS

QYZ

DHS

Table 2 Site location and long-term climate variables

Code CBS QYZ DHS

Location 42°24′N, 128°06′E

26°44′N, 115°04′E

23°10′N, 112°32′E

Terrain flat hill upland

Elevation (m) 736 100 300

Annual Precipitation (mm) 695 1485 1956

Average temperature ( )℃ 4.0 17.9 21.1

Canopy height (m) 26 11 17

Age (yr) 200 24 100

CBS: a broad-leaved and Korean pine mixed forest. QYZ: a young evergreen coniferous plantation. DHS: an evergreen conifer and broad-leaved mixed forest.

Results

0

10

20

30

40

50

Fre

quen

cy (

%)

Obs1

0

10

20

30

40

50Obs2

2 4 60

10

20

30

40

50

C1(× 10-3)

Obs3

1 2

C2(× 10 3- )

1 2

C3(× 10 4- )

1 2

C4(× 10 2- )

1 2

C5(× 10 3- )

3 4 5 6

C6(× 10 3- )

1 2

C7(× 10 4- )

2 4 6 8

C8(× 10 6- )

0

10

20

30

40Obs2

0

10

20

30

40

Fre

quen

cy(%

)

Obs1

1 20

10

20

30

40

C1(× 10 3- )

Obs3

1 2

C2(× 10 3- )

1 2

C3(× 10 4- )

1 2

C4(× 10 2- )

1 2

C5(× 10 3- )

3 4 5 6

C6(× 10 4- )

1 2

C7(× 10 4- )

2 4 6 8

C8(× 10 6- )

0

10

20

30Obs2

0

10

20

30

40

50

Fre

quen

cy(%

)

Obs1

1 20

10

20

30

40

c1(× 10 3- )

Obs3

1 2

c2(× 10 3- )

1 2

c3(× 10 4- )

1 2

c4(× 10 2- )

1 2

c5(× 10 3- )

3 4 5 6

c6(× 10 3- )

1 2

c7(× 10 4- )

2 4 6 8

c8(× 10 6- )

Our results showed that the estimates of parameters c1, c2 and c3 will not be influenced by NEE data, but constrained by carbon pools data. Because there is no CO2 release when carbon transfers from leaf, fine root and woody biomass to litter pool. On the contrary, carbon transfers between litter, microbes, slow SOM and passive SOM accompanied with CO2 release. Thus, the estimates of parameters c4, c5, c6, c7 and c8 changed when adding NEE to data sets.

Fig. 1 Posterior Distribution of parameters with 3 different assimilation experiments in CBS site

Fig. 2 Posterior Distribution of parameters with 3 different assimilation experiments in QYZ site

Fig. 3 Posterior Distribution of parameters with 3 different assimilation experiments in DHS site

AcknowledgementsThis work was supported by Chinese Academy of Sciences International Partnership Project "Human Activities and Ecosystem Changes" (No CXTD-Z2005-1). We thank all related staffs of ChinaFLUX and CERN for their contribution from observation to data processing.