Bayesian meta-analysis of regional biomass factors for Quercus mongolica forests in South Korea

Supplementary Material

Tzeng Yih Lam

Xiaodong Li

Rae Hyun Kim

Kyeong Hak Lee

Yeong Mo Son

S1. WinBUGS Codes for Biomass Conversion and Expansion Factor (BCEF)

The followings are WinBUGS codes for modeling regional level BCEF under multilevel Bayesian Hierarchical Model (BHM) framework assuming three Probability Density Functions (PDFs).

S1.1 Gamma PDF

Model{

# Random effects model

for(i in 1:N){

bcef[i]~dgamma(a.hat[i],b.hat[i])

a.hat[i]<-a[study[i]]

b.hat[i]<-b[study[i]]

}

# Prior distributions for Gamma distribution parameters

for(j in 1:J){

a[j]~dgamma(A.a,B.a)

b[j]~dgamma(A.b,B.b)

}

# Prior distribution for hyper-parameters

A.a~dgamma(0.001,0.001)

B.a~dgamma(0.001,0.001)

A.b~dgamma(0.001,0.001)

B.b~dgamma(0.001,0.001)

# Outputs

for(j in 1:J){

bcef.pub[j]<-a[j]/b[j]

bcef.pub.var[j]<-a[j]/pow(b[j],2)

}

post.a<-A.a/B.a

post.b<-A.b/B.b

bcef.mean<-post.a/post.b

bcef.var<-post.a/pow(post.b,2)

}

# Data

list(N=52,J=12,

bcef=c(0.3872671, 0.4534884, 0.5130233, 0.9871638, 1.3583333, 0.8590283,

0.8136239, 1.3797368, 0.6769231, 1.0243902, 0.7144781, 0.7646552,

0.9, 1.3829787, 1.264697, 2.8872222, 1.1297826, 1.2640756, 1.0467337,

0.8224576, 1.255814, 0.8483051, 1.8363636, 1.0052632, 0.9778761,

1.0751773, 0.743314, 0.615, 1.8353933, 1.0866972, 2.216, 1.0290244,

1.1546667, 0.5260759, 1.1733333, 0.9492105, 1.2691782, 0.7871034,

0.8320918, 0.8839029, 1.0295852, 1.1604706, 0.7522128, 0.6790374,

0.1865383, 0.8012882, 0.6543203, 0.7005553, 0.8231579, 0.9083227,

1.0767969, 1.0498184),

study=c(1, 1, 1, 2, 3, 4, 4, 4, 5, 5, 5, 5, 5, 5, 6, 6, 6, 6, 7, 7, 7, 7, 7,

7, 7, 7, 7, 7, 7, 7, 8, 8, 8, 9, 10, 11, 12, 12, 12, 12, 12, 12, 12,

12, 12, 12, 12, 12, 12, 12, 12, 12)

)

S1.2 Log-normal PDF

Model{

# Random effects model

for(i in 1:N){

bcef[i]~dlnorm(mu.hat[i],tau)

mu.hat[i]<-mu[study[i]]

}

# Prior distributions for Log-normal distribution parameters

for(j in 1:J){

mu[j]~dnorm(theta,tau.mu)

}

# Prior distribution for hyper-parameters

tau<-pow(sigma,-2)

sigma~dunif(0,100)

theta~dnorm(0,0.000001)

tau.mu<-pow(sigma.mu,-2)

sigma.mu~dunif(0,100)

# Outputs

inv.tau<-pow(tau,-1)

for(j in 1:J){

bcef.pub[j]<-exp(mu[j]+0.5*inv.tau)

bcef.pub.var[j]<-(exp(inv.tau)-1)*exp(2*mu[j]+inv.tau)

}

bcef.mean<-exp(theta+0.5*inv.tau)

bcef.var<-(exp(inv.tau)-1)*exp(2*theta+inv.tau)

}

# Data

list(N=52,J=12,

bcef=c(0.3872671, 0.4534884, 0.5130233, 0.9871638, 1.3583333, 0.8590283,

0.8136239, 1.3797368, 0.6769231, 1.0243902, 0.7144781, 0.7646552,

0.9, 1.3829787, 1.264697, 2.8872222, 1.1297826, 1.2640756, 1.0467337,

0.8224576, 1.255814, 0.8483051, 1.8363636, 1.0052632, 0.9778761,

1.0751773, 0.743314, 0.615, 1.8353933, 1.0866972, 2.216, 1.0290244,

1.1546667, 0.5260759, 1.1733333, 0.9492105, 1.2691782, 0.7871034,

0.8320918, 0.8839029, 1.0295852, 1.1604706, 0.7522128, 0.6790374,

0.1865383, 0.8012882, 0.6543203, 0.7005553, 0.8231579, 0.9083227,

1.0767969, 1.0498184),

study=c(1, 1, 1, 2, 3, 4, 4, 4, 5, 5, 5, 5, 5, 5, 6, 6, 6, 6, 7, 7, 7, 7, 7,

7, 7, 7, 7, 7, 7, 7, 8, 8, 8, 9, 10, 11, 12, 12, 12, 12, 12, 12, 12,

12, 12, 12, 12, 12, 12, 12, 12, 12)

)

S1.3 Weibull PDF

Model{

# Random effects model

for(i in 1:N){

bcef[i]~dweib(v.hat[i],lambda.hat[i])

v.hat[i]<-v[study[i]]

lambda.hat[i]<-lambda[study[i]]

}

# Prior distributions for Weibull distribution parameters

for(j in 1:J){

v[j]~dgamma(A.v,B.v)

lambda[j]~dgamma(A.lambda,B.lambda)

}

# Prior distribution for hyper-parameters

A.v~dgamma(0.001,0.001)

B.v~dgamma(0.001,0.001)

A.lambda~dgamma(0.001,0.001)

B.lambda~dgamma(0.001,0.001)

# Outputs

for(j in 1:J){

inv.v[j]<-pow(v[j],-1)

bcef.pub[j]<-pow(lambda[j],-inv.v[j])*exp(loggam(1+inv.v[j]))

bcef.pub.var[j]<-(exp(loggam(1+2*inv.v[j]))-pow(exp(loggam(1+inv.v[j])),2))*

pow(lambda[j],-2*inv.v[j])

}

post.lambda<-A.lambda/B.lambda

post.v<-A.v/B.v

inv.post.v<-pow(post.v,-1)

bcef.mean<-pow(post.lambda,-inv.post.v)*exp(loggam(1+inv.post.v))

bcef.var<-(exp(loggam(1+2*inv.post.v))-pow(exp(loggam(1+inv.post.v)),2))*

pow(post.lambda,-2*inv.post.v)

}

# Data

list(N=52,J=12,

bcef=c(0.3872671, 0.4534884, 0.5130233, 0.9871638, 1.3583333, 0.8590283,

0.8136239, 1.3797368, 0.6769231, 1.0243902, 0.7144781, 0.7646552,

0.9, 1.3829787, 1.264697, 2.8872222, 1.1297826, 1.2640756, 1.0467337,

0.8224576, 1.255814, 0.8483051, 1.8363636, 1.0052632, 0.9778761,

1.0751773, 0.743314, 0.615, 1.8353933, 1.0866972, 2.216, 1.0290244,

1.1546667, 0.5260759, 1.1733333, 0.9492105, 1.2691782, 0.7871034,

0.8320918, 0.8839029, 1.0295852, 1.1604706, 0.7522128, 0.6790374,

0.1865383, 0.8012882, 0.6543203, 0.7005553, 0.8231579, 0.9083227,

1.0767969, 1.0498184),

study=c(1, 1, 1, 2, 3, 4, 4, 4, 5, 5, 5, 5, 5, 5, 6, 6, 6, 6, 7, 7, 7, 7, 7,

7, 7, 7, 7, 7, 7, 7, 8, 8, 8, 9, 10, 11, 12, 12, 12, 12, 12, 12, 12,

12, 12, 12, 12, 12, 12, 12, 12, 12)

)

S2. WinBUGS Codes for Biomass Expansion Factor (BEF)

The followings are WinBUGS codes for modeling regional level BEF under multilevel Bayesian Hierarchical Model (BHM) framework assuming three Probability Density Functions (PDFs).

S2.1 Gamma PDF

Model{

# Random effects model

for(i in 1:N){

sbef[i]~dgamma(a.hat[i],b.hat[i])

a.hat[i]<-a[study[i]]

b.hat[i]<-b[study[i]]

}

# Prior distributions for Gamma distribution parameters

for(j in 1:J){

a[j]~dgamma(A.a,B.a)

b[j]~dgamma(A.b,B.b)

}

# Prior distribution for hyper-parameters

A.a~dgamma(0.001,0.001)

B.a~dgamma(0.001,0.001)

A.b~dgamma(0.001,0.001)

B.b~dgamma(0.001,0.001)

# Outputs

for(j in 1:J){

sbef.pub[j]<-a[j]/b[j]

sbef.pub.var[j]<-a[j]/pow(b[j],2)

}

post.a<-A.a/B.a

post.b<-A.b/B.b

sbef.mean<-post.a/post.b

sbef.var<-post.a/pow(post.b,2)

}

# Data

list(N=40,J=11,

sbef=c(0.4459647, 0.3270949, 0.7042645, 0.4194682, 0.3892045, 0.3785083,

0.2576757, 0.2809675, 0.3273002, 0.4334471, 0.3797139, 0.258156,

0.4315436, 0.1926606, 0.3566475, 0.3230652, 0.2885656, 0.2915897,

0.292745, 0.246013, 0.2433597, 0.1308844, 0.4380186, 0.7746617,

0.9052, 0.6962705, 0.2389068, 0.5028143, 0.1565736, 0.8610043,

0.5901667, 0.4062117, 0.4206047, 0.9849434, 0.3440617, 1.0287489,

0.8800643, 0.9973532, 0.6743402, 0.7825875),

study=c(1, 1, 1, 2, 3, 4, 4, 4, 5, 5, 5, 5, 5, 5, 6, 6, 6, 6, 7, 7, 7, 8, 9,

10, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11)

)

S2.2 Log-normal PDF

Model{

# Random effects model

for(i in 1:N){

sbef[i]~dlnorm(mu.hat[i],tau)

mu.hat[i]<-mu[study[i]]

}

# Prior distributions for Log-normal distribution parameters

for(j in 1:J){

mu[j]~dnorm(theta,tau.mu)

}

# Prior distribution for hyper-parameters

tau<-pow(sigma,-2)

sigma~dunif(0,100)

theta~dnorm(0,0.000001)

tau.mu<-pow(sigma.mu,-2)

sigma.mu~dunif(0,100)

# Outputs

inv.tau<-pow(tau,-1)

for(j in 1:J){

sbef.pub[j]<-exp(mu[j]+0.5*inv.tau)

sbef.pub.var[j]<-(exp(inv.tau)-1)*exp(2*mu[j]+inv.tau)

}

sbef.mean<-exp(theta+0.5*inv.tau)

sbef.var<-(exp(inv.tau)-1)*exp(2*theta+inv.tau)

}

# Data

list(N=40,J=11,

sbef=c(0.4459647, 0.3270949, 0.7042645, 0.4194682, 0.3892045, 0.3785083,

0.2576757, 0.2809675, 0.3273002, 0.4334471, 0.3797139, 0.258156,

0.4315436, 0.1926606, 0.3566475, 0.3230652, 0.2885656, 0.2915897,

0.292745, 0.246013, 0.2433597, 0.1308844, 0.4380186, 0.7746617,

0.9052, 0.6962705, 0.2389068, 0.5028143, 0.1565736, 0.8610043,

0.5901667, 0.4062117, 0.4206047, 0.9849434, 0.3440617, 1.0287489,

0.8800643, 0.9973532, 0.6743402, 0.7825875),

study=c(1, 1, 1, 2, 3, 4, 4, 4, 5, 5, 5, 5, 5, 5, 6, 6, 6, 6, 7, 7, 7, 8, 9,

10, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11)

)

S2.3 Weibull PDF

Model{

# Random effects model

for(i in 1:N){

sbef[i]~dweib(v.hat[i],lambda.hat[i])

v.hat[i]<-v[study[i]]

lambda.hat[i]<-lambda[study[i]]

}

# Prior distributions for Weibull distribution parameters

for(j in 1:J){

v[j]~dgamma(A.v,B.v)

lambda[j]~dgamma(A.lambda,B.lambda)

}

# Prior distribution for hyper-parameters

A.v~dgamma(0.001,0.001)

B.v~dgamma(0.001,0.001)

A.lambda~dgamma(0.001,0.001)

B.lambda~dgamma(0.001,0.001)

# Outputs

for(j in 1:J){

inv.v[j]<-pow(v[j],-1)

sbef.pub[j]<-pow(lambda[j],-inv.v[j])*exp(loggam(1+inv.v[j]))

sbef.pub.var[j]<-(exp(loggam(1+2*inv.v[j]))-pow(exp(loggam(1+inv.v[j])),2))*

pow(lambda[j],-2*inv.v[j])

}

post.lambda<-A.lambda/B.lambda

post.v<-A.v/B.v

inv.post.v<-pow(post.v,-1)

sbef.mean<-pow(post.lambda,-inv.post.v)*exp(loggam(1+inv.post.v))

sbef.var<-(exp(loggam(1+2*inv.post.v))-pow(exp(loggam(1+inv.post.v)),2))*

pow(post.lambda,-2*inv.post.v)

}

# Data

list(N=40,J=11,

sbef=c(0.4459647, 0.3270949, 0.7042645, 0.4194682, 0.3892045, 0.3785083,

0.2576757, 0.2809675, 0.3273002, 0.4334471, 0.3797139, 0.258156,

0.4315436, 0.1926606, 0.3566475, 0.3230652, 0.2885656, 0.2915897,

0.292745, 0.246013, 0.2433597, 0.1308844, 0.4380186, 0.7746617,

0.9052, 0.6962705, 0.2389068, 0.5028143, 0.1565736, 0.8610043,

0.5901667, 0.4062117, 0.4206047, 0.9849434, 0.3440617, 1.0287489,

0.8800643, 0.9973532, 0.6743402, 0.7825875),

study=c(1, 1, 1, 2, 3, 4, 4, 4, 5, 5, 5, 5, 5, 5, 6, 6, 6, 6, 7, 7, 7, 8, 9,

10, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11)

)

S3. WinBUGS Codes for Root-to-Shoot Ratio (RSR)

The followings are WinBUGS codes for modeling regional level R under multilevel Bayesian Hierarchical Model (BHM) framework assuming two Probability Density Functions (PDFs).

S3.1 Gamma PDF

Model{

# Random effects model

for(i in 1:N){

r[i]~dgamma(a.hat[i],b.hat[i])

a.hat[i]<-a[study[i]]

b.hat[i]<-b[study[i]]

}

# Prior distributions for Gamma distribution parameters

for(j in 1:J){

a[j]~dgamma(A.a,B.a)

b[j]~dgamma(A.b,B.b)

}

# Prior distribution for hyper-parameters

A.a~dgamma(0.001,0.001)

B.a~dgamma(0.001,0.001)

A.b~dgamma(0.001,0.001)

B.b~dgamma(0.001,0.001)

# Outputs

for(j in 1:J){

r.pub[j]<-a[j]/b[j]

r.pub.var[j]<-a[j]/pow(b[j],2)

}

post.a<-A.a/B.a

post.b<-A.b/B.b

r.mean<-post.a/post.b

r.var<-post.a/pow(post.b,2)

}

# Data

list(N=47,J=9,

r=c(0.2147239, 0.1919597, 0.196933, 0.2633352, 0.2022727, 0.1922619,

0.1918002, 0.1967306, 0.2034693, 0.1917949, 0.151831, 0.1636521,

0.181547, 0.2110022, 0.193951, 0.1849562, 0.1958333, 0.1998002,

0.1764176, 0.1874346, 0.2633484, 0.2631926, 0.1521314, 0.1645146,

0.1665136, 0.2401857, 0.1407942, 0.19586, 0.2048829, 0.3267565,

0.3140092, 0.1421093, 0.2064128, 0.0953246, 0.3292643, 0.325117,

1.1195047, 0.5079447, 0.1699677, 0.1456458, 0.9726821, 0.3164511,

1.0467703, 0.4526007, 0.7343642, 0.155599, 0.1425218),

study=c(1, 2, 2, 2, 3, 3, 3, 3, 3, 3, 4, 4, 4, 4, 5, 5, 5, 5, 5, 5, 5, 5, 5,

5, 5, 5, 6, 6, 6, 7, 8, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9,

9)

)

S3.2 Log-normal PDF

Model{

# Random effects model

for(i in 1:N){

r[i]~dlnorm(mu.hat[i],tau)

mu.hat[i]<-mu[study[i]]

}

# Prior distributions for Log-normal distribution parameters

for(j in 1:J){

mu[j]~dnorm(theta,tau.mu)

}

# Prior distribution for hyper-parameters

tau<-pow(sigma,-2)

sigma~dunif(0,100)

theta~dnorm(0,0.000001)

tau.mu<-pow(sigma.mu,-2)

sigma.mu~dunif(0,100)

# Outputs

inv.tau<-pow(tau,-1)

for(j in 1:J){

r.pub[j]<-exp(mu[j]+0.5*inv.tau)

r.pub.var[j]<-(exp(inv.tau)-1)*exp(2*mu[j]+inv.tau)

}

r.mean<-exp(theta+0.5*inv.tau)

r.var<-(exp(inv.tau)-1)*exp(2*theta+inv.tau)

}

# Data

list(N=47,J=9,

r=c(0.2147239, 0.1919597, 0.196933, 0.2633352, 0.2022727, 0.1922619,

0.1918002, 0.1967306, 0.2034693, 0.1917949, 0.151831, 0.1636521,

0.181547, 0.2110022, 0.193951, 0.1849562, 0.1958333, 0.1998002,

0.1764176, 0.1874346, 0.2633484, 0.2631926, 0.1521314, 0.1645146,

0.1665136, 0.2401857, 0.1407942, 0.19586, 0.2048829, 0.3267565,

0.3140092, 0.1421093, 0.2064128, 0.0953246, 0.3292643, 0.325117,

1.1195047, 0.5079447, 0.1699677, 0.1456458, 0.9726821, 0.3164511,

1.0467703, 0.4526007, 0.7343642, 0.155599, 0.1425218),

study=c(1, 2, 2, 2, 3, 3, 3, 3, 3, 3, 4, 4, 4, 4, 5, 5, 5, 5, 5, 5, 5, 5, 5,

5, 5, 5, 6, 6, 6, 7, 8, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9,

9)

)

Table

Table S1 Stem and above-ground biomass (ton/ha) for individual study site of each publication used for meta-analysis of BCEF, BEF and RSR.

Publication / Site No. / Stem Biomass
(ton/ha) / Aboveground Biomass (ton/ha)
Kwon and Lee 2006a / 1 / 132.6 / 176
Kwon and Lee 2006a / 2 / 117.2 / 168
Kwon and Lee 2006a / 3 / 153.8 / 212.2
Kwon and Lee 2006a / 4 / 141 / 177.4
Kwon and Lee 2006a / 5 / 149 / 213.3
Kwon and Lee 2006a / 6 / 163.5 / 195
Kwon and Lee 2006b / 1 / 184.58 / 250.41
Kwon and Lee 2006b / 2 / 157.12 / 207.88
Kwon and Lee 2006b / 3 / 241.99 / 311.82
Kwon and Lee 2006b / 4 / 232.93 / 300.85
Kwon and Lee 2006c / 1 / n/aa / 208.3
Kwon and Lee 2006c / 2 / n/aa / 194.1
Kwon and Lee 2006c / 3 / n/aa / 216
Kwon and Lee 2006c / 4 / n/aa / 200.2
Kwon and Lee 2006c / 5 / n/aa / 222.2
Kwon and Lee 2006c / 6 / n/aa / 191
Kwon and Lee 2006c / 7 / n/aa / 110.5
Kwon and Lee 2006c / 8 / n/aa / 151.6
Kwon and Lee 2006c / 9 / n/aa / 255.7
Kwon and Lee 2006c / 10 / n/aa / 209.1
Kwon and Lee 2006c / 11 / n/aa / 326.7
Kwon and Lee 2006c / 12 / n/aa / 236.9
Lee and Kwon 2006 / 1 / 153.92 / 212.18
Lee and Kwon 2006 / 2 / 141.03 / 177.37
Lee and Kwon 2006 / 3 / 122.79 / 157.29
Lee and Park 1987 / 1 / 78.33 / 112.64
Park 2003 / 1 / 43.12 / 62.35
Park 2003 / 2 / 117.55 / 156
Park 2003 / 3 / 64.72 / 110.3
Park and Moon 1994 / 1 / 70.4 / 97.8
Park et al. 1996 / 1 / 40.65 / 72.14
Park et al. 2005a / 1 / 188.56 / 243.76
Park et al. 2005a / 2 / 101.58 / 126.57
Park et al. 2005a / 3 / 97.51 / 121.24
Park et al. 2005b / 1 / 73.5 / 83.12
Song and Lee 1996 / 1 / 91.02 / 129.2
KFRI 2010 / 1 / 86.44 / 164.68
KFRI 2010 / 2 / 111.91 / 189.83
KFRI 2010 / 3 / 182.18 / 225.7
KFRI 2010 / 4 / 82.97 / 124.69
KFRI 2010 / 5 / 228.79 / 264.61
KFRI 2010 / 6 / 20.48 / 38.11
KFRI 2010 / 7 / 97.6 / 155.2
KFRI 2010 / 8 / 77.27 / 108.66
KFRI 2010 / 9 / 195.15 / 277.24
KFRI 2010 / 10 / 68.63 / 136.22
KFRI 2010 / 11 / 91.82 / 123.41
KFRI 2010 / 12 / 39.9 / 80.94
KFRI 2010 / 13 / 54.16 / 101.82
KFRI 2010 / 14 / 54.78 / 109.41
KFRI 2010 / 15 / 59.38 / 99.43
KFRI 2010 / 16 / 92.14 / 164.25

aStem biomass was not available.

Table S2 Median, 2.5% and 97.5% sample quantiles of hyperparameters for each fitted PDF for Biomass Conversion and Expansion Factor (BCEF).

Median / 2.5% Quantile / 97.5% Quantile
Gamma
Aα / 61.52 / 6.058 / 1522.0
Bα / 7.798 / 0.694 / 202.5
Aβ / 26.04 / 4.276 / 693.6
Bβ / 3.393 / 0.439 / 97.98
Log-normal
θ / -0.0570 / -0.2954 / 0.1854
φ / 11.18 / 2.731 / 132.7
Weibull
Av / 39.34 / 3.129 / 978.1
Bv / 13.0 / 0.8783 / 348.6
Aλ / 2.423 / 0.5187 / 111.9
Bλ / 2.837 / 0.3 / 144.0

Table S3 Median, 2.5% and 97.5% sample quantiles of hyperparameters for each PDF for Biomass Expansion Factor (BEF).

Median / 2.5% Quantile / 97.5% Quantile
Gamma
Aα / 93.52 / 7.208 / 1287.0
Bα / 15.45 / 1.006 / 219.4
Aβ / 11.55 / 2.319 / 261.7
Bβ / 0.7132 / 0.106 / 27.35
Log-normal
θ / -0.9846 / -1.29 / -0.6992
φ / 9.382 / 1.773 / 84.97
Weibull
Av / 95.16 / 6.333 / 1185.0
Bv / 28.98 / 1.568 / 379.7
Aλ / 1.064 / 0.3011 / 4.077
Bλ / 0.0393 / 0.0011 / 0.5047

Table S4 Median, 2.5% and 97.5% sample quantiles of hyperparameters for each PDF for Root-to-Shoot Ratio (RSR).

Median / 2.5% Quantile / 97.5% Quantile
Gamma
Aα / 1.342 / 0.397 / 4.577
Bα / 0.0393 / 0.0040 / 0.3034
Aβ / 1.1 / 0.3458 / 3.244
Bβ / 0.0070 / 0.0009 / 0.0451
Log-normal
θ / -1.484 / -1.736 / -1.242
φ / 24.99 / 3.293 / 2235.0